The robots are coming for your job!

We’re all headed for the breadline while the clankers steal all the jobs like a wave of filthy metal immigrants swarming our shores!

Soon it’s 50% unemployment or worse. 75% or 100%! Everyone is worthless in the age of AI. We need UBI now! The only thing left for everyone is living in the Stacks, Ready Player One style, eking out a sad existence in a dream world!

The AI jobs apocalypse is one of those problems that a lot of people really, truly believe is either already happening (it’s not) or that it’s inevitable (also not).

But it only exists in our imagination. It’s not real. It doesn’t have to become real and if history is any teacher, it won’t become real. But that doesn’t stop a steady drumbeat of people puffing their chest and proudly proclaiming we’ve got to act right now before it’s too late!

At this point, with even Bill Gates piling on, it’s become the Population Bomb of our time. Remember, the population bomb never happened. Instead, we got the Green Revolution and our population increased dramatically and there’s less hunger than ever in history.

But that didn’t stop Ehrlich and others from swearing it was real and coming soon. Ehrlich never recanted and believed it right up until his death a few years ago.

That’s the problem when we just assume the future is inevitable.

We can’t see the solutions hidden around the bend.

We make the mistake of thinking you just change one variable and everything else stays fixed. But that’s not how life works. Everything is changing everywhere, all at once.

And worse, we cause real problems trying to solve imaginary ones. Deep belief in the Population Bomb gave China the One Child Policy and millions upon millions of forced abortions and sterilizations and now an inevitable population decline in China that can’t be reversed this generation with a carrot or a stick.

But our fear of the jobs apocalypse is a distinctly western delusion this time. It’s the symptom of a deeper malaise in the West. We’ve lost the ability to see Star Trek futures or any positive future at all, or even realistic futures. Instead, we’re stuck in a forever info war of black and white reality takes. This or that. Doom or utopia!

But life is usually somewhere in between, with occasional swings to the extremes and usually never in the way we imagined.

The jobs apocalypse is a now a mass societal level hallucination. Its fundamental error is simple:

They assume a finite problem space.

They look at the economy and see a fixed amount of work to be done, a pie that can only be sliced thinner as machines take bigger bites. They see humans and machine competing for the same scarce resources. But the pie isn’t fixed. It never was. And the reason it isn’t fixed is baked into the very nature of technology itself.

Technology is nothing but abstraction stacking. And abstraction stacking is infinite. Therefore the work is infinite.

Let me show you what I mean.

Mud Huts and Hammers

Go back far enough and we lived in mud huts.

A mud hut is simple. You gather mud, you gather sticks, you pack it together, you live in it.

The problems associated with a mud hut are pretty simple and straightforward: keep the rain out, keep the animals out, don’t let it collapse on you while you sleep.

Now someone invents the hammer. And nails. And sawed boards.

These tools don’t just make building easier. They abstract away the old problem.

“How do I join two pieces of wood together?” used to be a problem that required skill, time, ingenuity. The hammer and nail automated that problem. They turned it into something trivially easy, something you don’t even think about anymore.

But here’s what happened next: because the joining problem was solved, we could think above it. We stopped worrying about how to connect pieces of wood and started thinking about architecture. About load-bearing walls. About second stories. About windows and doors and rooms with specific purposes instead of everyone jumbled together in a few rooms, a kitchen here, a bedroom there, a workshop out back, an upstairs and a downstairs.

The hammer didn’t reduce the amount of work. It moved the work up the stack. And the new work was more complex, more varied, and more interesting than the old work.

This is the pattern that repeats at every level of technological development, and it never, ever reverses.

Once you have houses instead of mud huts, you have a cascade of new challenges that didn’t exist before. Plumbing. Wiring. Insulation. Roofing materials that don’t rot. Drainage systems so the foundation doesn’t flood. Fire codes so your neighbor’s bad wiring doesn’t burn down the whole block.

Each of those problems becomes a job. A plumber. An electrician. An insulator. A roofer. A civil engineer. A building inspector. None of those jobs existed when we lived in mud huts. They exist because we solved the mud hut problem.

Now give us steel and concrete and suddenly we can build skyscrapers. And skyscrapers bring their own explosion of complexity: structural engineering, elevator systems, HVAC, fire suppression, seismic design, window washing at 800 feet, lobby security, electrical substations in the basement, telecommunications infrastructure on the roof.

Every solved problem is a platform for new problems. Every abstracted solution becomes the foundation for a new layer of complexity. The work doesn’t shrink. It compounds.

This is what I call the complexity ratchet. It only turns one way: toward more.

More problems, more solutions, more specialization, more jobs, more industries, more entire sectors of the economy that didn’t exist a generation ago.

The Abstraction Stack

When you understand this you can see all of human development as a stack of abstraction layers, each one built on top of the ones below it.

Humans are a special kind of animal.

Most other animals are still living exactly the way they lived 10,0000 years ago or a million years ago. An alligator swims and hunts food in the water. Gorillas form small groups that don’t get much bigger than 5 to 30 gorillas. Some massive ones may hit 60. But gorillas don’t really change much over time. They mostly live the same way they did in the distance past.

What makes us different is a few key things. We can learn from other humans and the world around us. We can record that knowledge and pass it on with language and writing. One person can solve a problem and others can learn how to solve it. Our knowledge spreads and compounds. We build on our own ideas and the ideas of others. We build on the solutions of the past.

A gorilla may make a stick tool to pry something out of a hole but it doesn’t get more complex than that. There are no other gorillas who look at that and say but what if I tied sticks together in this way to make a shelter? And once that gets figured out, no other gorilla looks at that shelter and thinks “how can I make that shelter even better?”

But that’s what we do. We don’t just react to the environment and problems. We solve problems and pass that knowledge on and the people who come after us inherit the solutions of the people who came before us. We can build on those solutions and adapt them and make them better (or worse).

At the lowest level of life and abstraction, it’s simple: raw survival. Finding food. Building shelter. Making fire. These are the base-layer problems. Solve them and you have a new base layer of new kinds of problems to solve.

Each major wave of technology solved a base-layer problem and in doing so created an entirely new layer of problems above it:

· Agriculture solved “how do we reliably eat?” — and created problems of land ownership, irrigation, crop rotation, storage, trade, taxation, and governance.

· Writing solved “how do we remember things across generations?” — and created problems of literacy, education, record-keeping, law, bureaucracy, and literature.

· The printing press solved “how do we spread knowledge at scale?” — and created problems of intellectual property, censorship, journalism, publishing, public opinion, and democratic discourse.

· The steam engine solved “how do we generate mechanical power without muscles?” — and created problems of factory design, worker safety, urban planning, railroad engineering, coal mining, labor relations, and environmental pollution.

· Electricity solved “how do we deliver energy anywhere?” — and created problems of grid design, power generation, appliance manufacturing, electrical safety codes, utility regulation, and an entire consumer electronics industry.

· The Internet solved “how do we connect all human knowledge?” — and created problems of cybersecurity, digital privacy, online commerce, content moderation, network infrastructure, cloud computing, social media dynamics, and an entire digital economy that employs tens of millions.

Notice the pattern? Each solution didn’t just solve a single problem.

It created an entirely new problem space that was larger, more complex, and more varied than the one it replaced.

The stack grows. It never shrinks.

If you understand the abstraction stack concept, you understand why the work never stops.

Problems are infinite because solutions always create new problems.

Every time we solve something, we create new complexity. Complexity breeds complexity. New complexity means new needs. New needs mean new work. The universe of things-that-need-doing is not a fixed pool that machines are draining. It’s a river that gets wider and deeper with every damn we build.

No intelligence of any kind, human or machine can solve all the problems, because the very act of solving problems generates more problems. It’s not a bug. It’s the fundamental structure of progress.

Think about it this way: Did the invention of the automobile put everyone out of work?

It destroyed the horse-and-buggy industry, sure.

Farriers, stable boys, carriage makers, horse breeders, those jobs shrank or vanished.

But the automobile created: mechanics, gas station attendants, highway engineers, traffic cops, driving instructors, auto insurance agents, car salespeople, parking garage operators, car wash workers, trucking companies, suburban developers, shopping mall architects, fast food drive-throughs, road trip motels, auto parts stores, NASCAR drivers, Uber drivers, and eventually the entire modern logistics industry that moves every product you’ve ever bought.

The automobile didn’t just create more jobs than it destroyed. It created more kinds of jobs. More variety. More specialization. More layers of the stack.

AI is the same thing. It’s the next big abstraction layer. It will solve a current set of problems, so that we can move up the stack to the next layer. And that next layer will be more complex, more varied than the one we’re leaving behind.

But what kind of jobs says the job doomer? I need to know now and if you can’t think of all the jobs in the future then there must not be any!

Wrong.

The problem is we can’t fully see those new jobs that will come in the future but that doesn’t mean they won’t materialize. We can make some educated guesses but we’re always limited.

You can’t explain the job of a “web designer” to an 18th century farmer, because it’s built on a stack of solutions and challenges that don’t exist yet. Without electricity and the internet and computers you can’t even begin to imagine the concept of a web designer.

When you’re standing mid-way up a mountain, you can’t see the top until you climb up higher.

Way too often people make the mistake of seeing what will disappear or change. So they make the mistake that there is no future or that all the jobs will disappear and there won’t be anything to replace that work.

But it’s wrong. Just because you can’t see it doesn’t mean it doesn’t exist.

If you understand the mechanism of the infinite abstraction stack then you understand the work is endless and you develop a learned trust about the future.

The farrier in 1890 couldn’t imagine “NASCAR driver” or “Uber product manager” but eventually that work did exist. The telephone operator in 1950 couldn’t imagine “social media content strategist” of 2026. The typesetter in 1980 couldn’t imagine “UX designer.” The new jobs are invisible from below because they exist in a layer of the stack that hasn’t been built yet.

But they’ll be built. They always are. Because the problems are infinite.

And so are the solutions.

The Compounding Error

If you misunderstand this, if you get the basic understanding of the abstraction stack wrong, it corrupts all your understanding beyond that point.

If you start from the premise that work is finite, that there’s a fixed amount of stuff to be done and machines are eating through it, then every conclusion you draw from that premise will be wrong.

You’ll predict mass unemployment. You’ll demand UBI as the only possible response. You’ll want to ban or throttle AI development. You’ll see a zero-sum world where every machine gain is a human loss. You’ll build policy around scarcity when the reality is abundance.

And you’ll be wrong about all of it. Not because you’re stupid, many of the smartest people in the world make this mistake, but because the premise is wrong. The error compounds and radiates outward, corrupting every inference, every prediction, every policy recommendation.

It’s like building a skyscraper on a foundation that’s three degrees off true. At the ground floor, you can barely tell. By the fiftieth floor, the building is leaning so far it’s about to collapse. That’s what happens when you get the foundation wrong on this question.

(Source: WikiMedia Commons)

The lump of labor fallacy is the economic version of this error. But the abstraction stack goes deeper, it’s about the nature of problem-solving itself, the structure of progress, the reason technology creates more work rather than less. It’s the metaphysical argument for why the doomers are wrong.

Understand this and you understand the future. Understand that technology is abstraction stacking, that each layer creates more complexity than it resolves, that problems are the raw material of work and that the raw material is infinite. Understand that and the panic about AI evaporates.

Not because there’s nothing to worry about. There’s plenty to worry about: like privacy, surveillance, and weapons.

But the “AI will take all the jobs” fear? That’s not just wrong. It’s the opposite of how progress works.

The jobs are endless because the problems are endless. And the problems are endless because every solution is a platform for a hundred new problems.

That’s not a theory. That’s the history of civilization.

And AI doesn’t break the pattern. It extends it.

But What If AI Is Better at Everything?

This is one of those objections that people make that sounds like a checkmate.

Even people who nod along with the Infinite Stack will stop you here and say: “Okay, fine, the problems are infinite. But what if AI is just better at all of them? What if it’s better at solving problems, better at creating, better at building, better at everything humans do? Then why would anyone hire a human for anything?”

It’s a fair question. So let’s take it seriously.

First: it’s an assumption, not an actual fact.

AI might get better at everything. It might not. We’re extrapolating about a technology, Large Language Models, that have been in wide public use since 2020, so about six years at the time of this article’s writing, and treating it like a settled cosmic law.

Of course, Artificial Intelligence and machine learning has been around a lot longer than that, of course, in various forms. But when most people think of AI today they think of LLMs like GPT or Claude.

Experiments with language models existed as far back as the 1950s and 1960s. IBM and Georgetown University demonstrated one of the earliest machine learning translation systems in the 1954. It converted Russian phrases to English.

The first chat bot was created by an MIT researcher in 1966, ELIZA. It used simple pattern matching and basically tried to simulate conversation by reflecting people’s statements back as questions. It was primitive and didn’t work that well but it laid the ground work for the more human like interactions that we’re used to today.

The history of artificial intelligence is filled with wild overestimations regarding how quickly machines would achieve human-like intelligence and automate the workforce out of existence.

In 1984, Nils Nilsson, a Stanford AI Pioneer, who led the development of Shakey the Robot (the first mobile robot able to reason about its actions, wrote:

“We are beginning a transition period to a workless society... Within the next few decades, AI will be able to do all the jobs that humans are paid to do.”

Nope.

In 2013 a breathless Oxford University Study got regurgitated ad nauseum in the media:
Researchers famously predicted that 47% of all US jobs were at high risk of being completely automated by computers within “a decade or two.”

The Reality: A decade later, US unemployment reached historic lows.

In 2023, Dario Amodei (Anthropic CEO) warned that AI could wipe out up to half of all entry-level white-collar jobs and push US unemployment to 20% and as much as 50% within 1 to 5 years.

Three years later, Anthropic and everyone else is still hiring and we’re seeing a boom in programming jobs, the most AI effected job on the planet.

And now we have LLMs advanced in leaps and bounds. It seems every month we get a newer, more powerful model, capable of more and more general capabilities from creating images and video, to understanding images, to writing and understanding complex documents to writing computer code and much much more.

So maybe LLMs really are the tipping point, the fundamental architecture that causes intelligence to grow faster and faster. But don’t bet on it just yet.

Many of the top experts in the field see tremendous flaws and limits in the architecture and just making them bigger doesn’t magically solve those problems.

They don’t have real long terms memory for instance, and they can’t update what they learn the way we do every single day of our lives.

We pick up a hot cup the wrong way, burn our hand and learn never to do that again. We’ve adjusted our understanding. LLMs need to be retrained to learn something new and that takes time. It doesn’t happen on the fly.

The list of limitations of LLMs is long and varied. I won’t touch on all of them here but I’ll cover more of them as the book goes on, to give a peek into the future.

I can’t tell you exactly the kind of architecture we’d need to have truly revolutionary intelligent machines but I can give you some strong possibilities that the top researchers in the world are betting on right now. Not all of those research angles will work out. And we may hit some kind of limitation that we can’t see and can’t overcome without new kinds of hardware or architectures or algorithmic breakthroughs.

Or maybe attention is all you need and we are at a new frontier and truly astonishing intelligence is only a few years away.

The American labs are betting everything on the Big Bang of intelligence, where a sufficiently advanced model can design the successor to the Transformer and solve continual learning and long term memory and myriad other problems. Maybe it works? Maybe we get an exponential take off and exponential increase in intelligence? It’s a strong possibility but it’s also likely to hit hidden walls and turn out to be a hidden S curve. But it’s just one possibility. Treating it like dogma is a fundamental mistake. The future has a tendency to arrive when it feels like arriving, not when we predict that it will.

People said the same thing about expert systems in the 1980s. That intelligence was soon going exponential and robots would be everywhere. They said it about IBM’s Deep Blue in 1997 when it beat Gary Kasperov, the world champion, in chess. Chess was considered the ultimate intellectual pursuit. Turns out you didn’t need much real intelligence and you could build a machine to brute force try all the possibilities in the way a human couldn’t. They said the same thing about Watson in 2011 when it beat the world Jeopardy champions in answering questions. Each time, the technology was genuinely impressive, and each time, the extrapolation to “it’ll be better at everything soon” turned out to be wildly premature.

(Source: WikiMedia Commons)

Maybe this time really is different. Maybe large language models and their descendants really do scale to superhuman performance across every domain. I’m not ruling it out. But I’m also not going to build my entire worldview on a prediction that has a 0% historical success rate.

But here’s the thing that matters: even if you grant the strongest version of the claim, even if you assume AI becomes better than humans at every measurable task, it still doesn’t lead where the doomers think it leads.

And there are three reasons why.

Reason 1: Better at Everything Doesn’t Mean Should Do Everything

There’s an old idea in economics called comparative advantage, and it’s one of those concepts that’s so simple it sounds wrong until you think about it for ten minutes.

Here’s how it works. Let’s say AI is 1,000 times better than a human at drug discovery and only twice as good at comforting a grieving patient. In absolute terms, AI is “better” at both. But the efficient allocation is obvious: you want the AI doing drug discovery, where its advantage is enormous, and the human doing the comforting, where the gap is small and the human brings something the machine doesn’t, human connection and warmth.

It doesn’t matter that the AI could technically do the comforting “better” by some metric. What matters is opportunity cost. Every hour the AI spends comforting a patient is an hour it’s not spending discovering a drug that could save ten thousand lives. The math pushes the AI toward the tasks where its advantage is greatest, which means humans get pushed toward everything else, and “everything else” turns out to be a very large category.

David Ricardo figured this out in 1817 with countries and cloth and wine when he wrote and published “On the Principles of Political Economy and Taxation” that gave us the idea of comparative advantage.

He used a thought experiment with two countries, England and Portugal, producing two goods: cloth and wine.

Here’s the setup: Portugal is better at making both cloth and wine than England is. Portugal has what economists call an “absolute advantage” in everything. So naively, you’d think Portugal should just make everything itself and England is screwed.

But Ricardo showed that’s wrong. What matters isn’t who’s better, it’s who gives up less by specializing.

Portugal is a lot better at wine and only somewhat better at cloth. So Portugal’s opportunity cost of making cloth is high, every hour it spends on cloth is an hour it’s not spending on wine, where its advantage is enormous. England, meanwhile, is less bad at cloth than it is at wine. So England’s comparative advantage is in cloth.

If Portugal specializes in wine and England specializes in cloth and they trade, both countries end up with more of both goods than if they’d each tried to do everything themselves. Even though Portugal is better at everything.

It applies to humans and AI just as well.

Noah Smith wrote a detailed piece on this in his Noahpinion blog called “Plentiful, High-Paying Jobs in the Age of AI“, and the key insight is that comparative advantage works as long as there are any constraints on AI, compute costs, energy, deployment speed, regulatory limits, the fact that you can’t run infinite instances simultaneously.

And there are always constraints. There will always be constraints. Physics doesn’t hand out free lunches, not even to neural networks.

So “better at everything” doesn’t mean “should do everything.”

It means the work gets allocated differently. Humans shift toward the tasks where the gap is smallest and where human qualities, judgment, trust, accountability, embodied presence, emotional resonance, matter most.

Reason 2: We Want Humans to Do Some Things

This is the part the purely economic argument misses, and it might be the most important part.

Even if a robot could cook a better steak than any human chef on Earth, would you want to eat at a restaurant where every dish was made by a machine and served by a bot?

Some people would. Fair enough. But a lot of people wouldn’t.

And the reason isn’t rational in the economic sense. It’s human. We go to restaurants for the experience. For the human connection. For the story of the chef who spent twenty years perfecting her mole recipe because her grandmother taught her and she’s been chasing that exact flavor ever since. We want to see a face when we walk in. We want the waiter who recommends the wine because he actually tasted it and loved it, not because an algorithm scored it 0.3% higher on a flavor profile.

The data backs this up. The craft beer market hit $28.9 billion in 2023 in the U.S. alone and is projected to nearly triple globally by the mid-2030s, despite the fact that mass-produced beer is cheaper, more consistent, and by many objective measures “better” at being beer. People pay a premium for craft beer because it’s human-made. Because it has a story. Because the brewer is a person in their community who makes small batches with weird ingredients and gives them funny names.

Etsy exists because people can, do, and will pay more for handmade goods. Live concert revenue is growing faster than streaming revenue. The live music market hit $23.6 billion in 2024, projected to reach $40.6 billion by 2032, with an 8.8% growth rate that outpaces streaming’s slowing 5.9%. When recorded music became essentially free, the human live experience became more premium, not less.

That’s not a fluke. That’s a clear pattern.

Research on consumer preferences consistently shows that people value human made products more highly in creative, emotional, and subjective domains. They often show what psychologists call “algorithm aversion,” a preference for human judgment in morally charged decisions, in artistic expression, in anything where authenticity and story seem to matter most. The “human-made premium” is a real, documented, and growing economic phenomenon.

So here’s what I think actually happens in a world where AI is technically better at most things: the market shards and segments. You get cheap, fast, abundant machine-made goods and content for the utilitarian stuff, and you get a growing premium tier for human-made, human-delivered, human-authenticated work.

The robot cleans the grill and scrubs the toilet at the restaurant. The human face greets you at the door. The chef’s passion is on the plate. The waiter tells you about the wine. And you pay for it because that experience, that human experience, is what you actually came for.

Of course, other kinds of restaurants will exist too. Totally bot run. Cheap. Fast. You go in precisely because you don’t want to interact with anyone. You just want a lightning fast meal.

Both will exist and both have their place in society.

There are countless examples of this in the world today and they will only grow tomorrow.

We will want our stories told by people, because we relate to people.

We might accept human-machine collaborations, a filmmaker who uses AI tools to execute a vision that’s authentically hers, but we’ll likely reject the fully machine-generated story. At least some will. Some will embrace that too and that’s fine because it’s a choice we all get to make personally and one person’s choice doesn’t make your choice too.

We likely won’t reject the pure algorithm produced piece because it’s technically worse. We might reject it because we feel it has no soul, no passion, no struggle behind it.

Whether that is true or not doesn’t matter in the least.

Humans emotions are not rational. And that’s fine.

Because nobody suffered for that piece, nobody poured their heartbreak into it, nobody stayed up until 3 AM rewriting the ending because it didn’t feel true yet.

And if we can’t tell the difference? Well, we’ll find ways to verify it, the same way we verify organic food and fair-trade coffee. The provenance becomes part of the product.

Though I suspect people will find room for pure AI creativity too, the way we found room for electronic music alongside acoustic, for digital art alongside oil paintings, for e-books alongside hardcovers. Humans are big-tent creatures.

We don’t replace. We add.

Reason 3: The Jobs of the Future Are Invisible From Here

This is the argument that sounds like a cop-out until you realize it’s the most historically bulletproof claim in the entire article.

Ask an 18th century farmer to predict the job “elevator repair technician.” They can’t.

Again, not because they’re stupid, but because the job sits on top of a stack of inventions that don’t exist yet: steel-frame construction, electric motors, safety brakes, urban density economics, building codes, the entire concept of a skyscraper.

The job is invisible from below because the layer of the stack it lives on hasn’t been built yet.

Ask someone in 1900 to predict “air traffic controller.” They’d need to first invent airplanes, radar, radio communications, airports, and international aviation governance systems.

Ask someone in 1975 to predict “cloud infrastructure engineer.” They’d need personal computers, the Internet, server virtualization, distributed computing, and a global market for on-demand compute.

Ask someone in 2000 to predict “TikTok content strategist.” They’d need smartphones, 4G networks, algorithmic content feeds, a creator monetization ecosystem, and the entire cultural infrastructure of short-form video.

These aren’t just new labels for old jobs. They’re genuinely new work that requires capabilities and knowledge that didn’t exist in any form before the stack was built beneath them. David Autor at MIT calls this “new work” in his paper “What Makes New Work Different from More Work“ not just more work, but work that “imbues labor with new scarcity value” and “requires capabilities that are neither already abundantly available nor simply more of the same.”

And this is why the “AI is better at everything” argument falls apart at the deepest level. It assumes a fixed task list. It looks at the jobs that exist right now and asks “can AI do these better?” But the future economy won’t be organized around today’s task list any more than today’s economy is organized around the task list of 1900. The jobs of 2050 will be as unimaginable to us as “drone cinematographer” or “AI personality auditor” would have been to someone in 1980. We can’t name them because they live in a layer of the stack that hasn’t been built yet.

What we can do is foresee the kinds of problems that scale and scope will create, and from those problems, project the shadows and potential outlines of the jobs to come.

The Problems We Can See Coming

Let’s take something like a factory as an example. Right now in China we’re seeing the rise of “dark factories.” No they’re not making dark magic. They’re dark because no human works there. They don’t need to turn on the lights at all because there are no humans on the factory floor.

Now if factories go more and more dark (they will), let’s say If factories go dark, 90% machines, no lights because there are no humans who need them, Xiaomi already producing one smartphone per second with zero humans on the floor, that doesn’t mean zero jobs. It means the jobs shift. Robotics maintenance. Process engineering. Quality systems. Logistics orchestration. Compliance. Customer-specific design. The factory floor is dark but the control room, the design studio, and the customer operations center are lit up and hiring.

And dark factories create something genuinely new: the ability to dial production up and down algorithmically without mass layoffs. You couldn’t do that before without eliminating a lot of people.

Now you adjust machine schedules, reroute orders, alter software parameters. Demand drops? You scale down production without firing anyone, because there’s almost nobody on the floor to fire.

Demand surges? You ramp up without a hiring sprint.

That’s a fundamentally different economic structure, and it creates its own cascade of new problems to solve: demand forecasting, algorithmic supply management, dynamic pricing, waste prevention, environmental compliance for high-throughput automated facilities.

Dark factories also enable reshoring.

When labor costs are no longer the dominant factor, because the factory is 90% machines, suddenly it makes more sense to produce close to your customers rather than shipping everything across the Pacific. Bain & Company, in a report called “Microfactories: The Back-to-Local Moment“ calls it the “back-to-local moment,” and they explicitly compare it to the craft beer revolution: just as craft breweries emerged as agile local producers that challenged the big consolidated beer manufacturers, micro-factories can produce locally, flexibly, and at scales that were uneconomical before automation made them viable. They write “Are we in a back-to-local moment, with new automation allowing for much smaller factories making more bespoke products?”

The micro-factory market is growing at roughly 18% annually. That’s a new industry creating new jobs that didn’t exist five years ago.

Of course, Americans might not do this because they’re fighting re-industrialization with anti-data center mobs but maybe cooler heads prevail and we get robot factories in the US.

And if most goods become cheap and ubiquitous, not Star Trek replicator-level, but dramatically cheaper than today because the factories are dark and the supply chains are automated, then you get a proto-abundance economy. And abundance doesn’t eliminate economics. It migrates the scarcity.

When goods are cheap, what becomes scarce?

Attention. Trust. Human connection. Taste. Curation. Beautiful locations. Low-conflict digital or in-persona communities. Experiences you can’t download. Positional goods that signal identity.

The work shifts toward managing, creating, and providing those scarce things.

Every one of those scarcities is a job category waiting to be named.

What kinds of problems does algorithmic supply and demand create?

Overproduction and waste, because when making things is cheap, the temptation to flood the market is enormous.

Environmental externalities, because cheap production still uses materials and energy. Attention scarcity, because consumers can’t process infinite choice.

Market power concentration, because whoever controls the automation stack might control the economy.

Inequality, because cheap goods don’t guarantee broad ownership of productive capital.

Meaning and status competition, because when the basics are abundant, people compete harder on relative goods.

Every single one of those problems is work. Every single one of them requires human judgment, human values, human institutions, human governance.

The machines can make the stuff. But deciding what to make, how much to make, who gets it, what the environmental limits should be, how to prevent monopoly abuse, how to give people meaning when the old sources of meaning change:

That’s human work.

The future isn’t man working solo, any more than the agricultural era was man working by himself. That was man and animal muscle power working side by side. The industrial era was man and steam working hand and hand.

And now the AI era is a new kind of machine, the intelligent machine. Mind and machine working together, building more, faster and faster, expanding solutions and creating new problems to solve.

That’s the infinite stack, the never-ending cycle of new challenges and new work.

And if we enter a world of distributed intelligence, like in Star Wars and Star Trek, not centralized, gated intelligence, it gets even better. Then intelligence can evolve agency and intricacy and adaptations to different environments.

(Source: Lucas Films, Star Wars)

Think hyper-specialization.

R2D2 didn’t call the centralize Claude mind to run. Its scrappy little mind ran on its own hardware and evolved there too. R2D2 is a continual learner. A machine personality and a friend and radically aligned to people and its own unerring sense of heroic justice.

The Enterprise doesn’t send token requests off to Starfleet center to run the ship intelligence. It just runs on the local hardware.

The future is distributed. It’s full of adaption and evolution of intelligence. It’s full of specialization. It’s full of solved problems and new kinds of problems.

We’ve seen leaders of AI now promising to solve all disease in ten years. But we won’t. We’ll absolutely get meaningful new ways to fight the scourge of disease, but disease is a cat and mouse game and we’re up against biological systems so complex that a single protein’s machinery makes our most advanced computers look like calculators. It would take the ability to perfectly simulate the human body to answers questions like will this drug have side effects in 25 years? Right now the only way to solve it is to wait 25 years. We have algorithms that approximate reality and help us get answers faster and AI will give us many more of those but it won’t get us to Utopia.

Utopia means no place. Life is more complex. It’s something in between the extremes.

So we probably won’t solve all disease in a decade.

But that shouldn’t stop us.

We’ll make big progress. We’ll get closer to the Tricorder. We conquer previously unconquerable cancers and viruses.

To do that though, we have to keep pushing, keep trying, keep building, keep bolding going where nobody has gone before.

We have to believe in a better tomorrow.

That’s the biggest problem in the West right now. We’re fighting data centers of non-existent water consumption and we’ve got a 30+% approval rating of AI, while the Chinese have 80+%. They’re building robots and drone food deliver and autonomous trains while we’re turning away from the future and towards a past that’s already gone and won’t come back no matter how much we yearn for it.

We’ve got to embrace the future and trust in the endless and infinite stack that’s always there and always creating new mountains for us to climb.

The work will always grow. The challenges will always grow. We climb one mountain but there’s always another one.

And the work goes on, forever and ever.

To infinity.

Is anyone doing real work with AI agents?

AI agents can do almost anything. And that’s exactly why most people have no idea what to do with them.

They’re an infinite canvas, a wide open road. Where do you start? Anywhere. You just talk to the agent and ask and iterate and figure it out as you go.

But it doesn’t feel to me like most people are doing this.

I had a friend ask me “How do I learn how to use agents?”

And I just thought, “why would anyone even ask this? You just talk to it and figure it out.” You talk to it and ask and then keep iterating. That’s it.

It’s a bit like this meme:

The truth is most people don’t think “I’ll just try and fail and then adjust until it works.” They don’t have a backlog of a thousand projects they always wanted to do. They don’t have a million things they want to get done. They just want to know how to do one thing that is already worked out for them. And that’s not how agents work today.

But if you’re terminally online the way I am, people do seem to be doing something with agents, at least if you believe X and YouTube.

They’re running dozens of worktrees at the same time or snatching up 10K computers to run agents locally. NVIDIA just put out a workstation that costs 100K to run even bigger models. People will buy them.

YouTube is saturated with influencer videos about agents shouting about how “this is INSANE!”.

OpenAI and Anthropic are the fastest-growing companies in history. They make the British East India Company look tiny. They’ve got billions of users paying them massive amounts of money. Companies are burning through their yearly AI budgets in months.

OpenClaw and Hermes have hundreds of thousands of stars and millions of downloads. Codex went from under a million users to 20 million in months.

Grok Bot is hot right now and promises to make inroads with regular folks. But only time will tell if it’s easy enough to be the Instagram moment for agents.

But I ask you. If you really dig deeper and the question is:

“Is anyone actually doing any real work at all with these things?”

The answer is yeah, a few people. But mostly no. So I started to ask why? And as I dug deeper and deeper what I found is that too often it feels like working with agents is a sys-admin exercise. Every video on X and YouTube is basically:

Here’s how to set them up in 100 easy steps!

Register this API key. And then this one. Set up this system. Check it. Install twenty plugins. Check. Fix. Run the doctor. Troubleshoot. Do it again.

It’s a bit like running a marathon. You do it to say you ran the marathon, but what’s actually at the end? What’s the thing you were running towards or were you just running to run?

I’ve got dozens of agents running in parallel! Okay but what are they doing? The point seems to be getting them running in parallel instead of having an actual work task that requires agents running in parallel.

What’s being done, other than coding and making more agents?

And the answer for most folks is: not much.

I know how I use agents. Research. A writing sparring partner. Editing. Website creation and deployment. Checking the background of an accountant. Researching travel. Language tutor. Coding. DevOps work.

But mostly I what I found from other people is that they’re spending ninety percent of the time setting up the agents instead of doing anything real with them.

Early in 2026, Will Mandis wrote about this phenomenon in a scathing article called “Tool Shaped Objects” where he talks about a stunningly beautiful Japanese wood shaver from the 1700s, called a Kanna, that still gets made today. Its blades are forged from the finest steel and it takes dedicated and skilled craftsmen months to make one and set it up.

It’s a fine machine and it shaves wood with unmatched precision. It’s gorgeous, a machine work of art.

There’s just one problem.

“It is beautiful. It is also, in the economic sense, worthless. A power planer does the same work in a fraction of the time. The Kanna exists so that the setup can exist.”

So what does all this mean? Where is this all going if all we’re doing is a painful exercise of setup for setup’s sake?

Some folks see the fall of Babylon. Ed Zitron is no fan of AI and his writing practically salivates at the image of AI burning like Byzantium after its walls come crashing down. He digs into the scary numbers behind massive hyperscaler spending on AI and the spending is eye-watering:

“Microsoft has dedicated its Fairwater data centers (however much actually exists) entirely to OpenAI, much like Amazon has for Anthropic with however much of its massive Indiana-based Project Rainier has actually been turned on, and much like Google is in talks to backstop a $15 billion data center project for Anthropic, along with data centers with Cipher Mining and TeraWulf and a $35 billion private credit-funded Broadcom-backstopped deal where Google will sell Anthropic its TPU AI chips, put them in a Google-built data center, and rent them back to Anthropic.”

Yeah, but that’s a lot of demand obviously, right?

Maybe.

He goes on:

“I want to spell this out: when you remove Anthropic and OpenAI’s compute spend, I am not confident that Google, Microsoft and Amazon have much of an AI business.

While many people believe — largely because the big three refuse to break out their actual AI revenues or disclose their customer concentration — that they have AI revenues coming from a diverse set of different customers, the reality is that their largest cloud customers, let alone AI customers, are two companies that can literally not afford to pay them without a near-infinite flow of venture capital or debt.”

That is damning. Basically two companies driving all demand.

But Zitron seems to miss that those two companies mean massive demand from other companies, big and small, for those models. If nobody is using it then why are those companies growing faster than any other companies in history? Not just in revenue but in subscribers and API users?

So someone is using AI. But the question is, who? Like who is really doing anything with it, if at all?

And that’s where I’ve got some real nagging doubt.

I’ve been wondering whether anyone outside coding, the AI industry itself, and marketing is actually using agents to do real work? And I don’t mean asking an agent to rewrite an email. Or generating a picture of a cat dressed like Napoleon. I mean handing meaningful parts of your life and work to a machine and letting it help you do stuff autonomously.

At first I thought the problem was mostly technical. Agents are unreliable. They hallucinate. They need babysitting. Setting one up still feels like a sys-admin certification exam designed by a committee of sadists.

But that I realized it was something much deeper.

The Agency Problem

Most people absolutely cannot, or will not, use agents in their current form.

But it’s worse than that.

The big labs keep warning that AI is coming for everyone’s jobs. America’s doomer movement talks about the machines as if Skynet is already filling out the paperwork. NIMBY coalitions fight data centers over water use while happily golfing and munching burgers and posting about it on their phone on social media, never noticing the irony. Gen Z and Gen Alpha workers quietly sabotage AI productivity mandates with feigned acceptance and performative use because helping the machine do more of their job feels a lot like volunteering to erase their own future.

This is not the Luddites smashing factory machines with hammers. It is subtler, quieter, more passive resistance. People sandbag the rollout. They use the approved tool for the approved little task, paste the result into a document, and change nothing fundamental about how they work. They comply with AI without ever really adopting it.

My wife is smart and perfectly capable of using GPT. She’ll ask it for travel recommendations or lists. She once used it to help write stories while she was learning a new language. She also used an online AI tool that translates TV shows to help her learn, and that felt a little like magic. Useful, real things. But it hasn’t become her interface to the world. It is still a tool she visits occasionally, not a collaborator woven deeply into her day. It’s like taking a two week vacation, not moving to a new country.

And honestly, why would it be? The dominant interface is an empty box staring back at you.

It’s just blinking at you.

People ask me, “What can you do with AI?” My answer is: anything you can imagine.

That sounds liberating. In practice it is terrifyingly vague.

For high-agency people, “anything” is rocket fuel.

They have a million projects they have always wanted to try, companies they want to build, questions they want answered, trips they want planned, systems they want fixed, books they want written. Their problem is not finding a use for AI. For people like me, the problem is that we can’t do all of it fast enough.

But most people do not walk around with a private backlog of impossible dreams. Their job is to do their job and go home. Their job is usually to be told what needs doing, master the process, complete it, and stop thinking about it when the day ends.

Give them an infinite canvas and they don’t see freedom. They see another task: invent the task.

Most people aren’t imagining anything at all. They don’t need or want to do anything more. Give them the ability to do anything and they’d do absolutely nothing and be perfectly happy about it.

Most people dream of the beach and living by the beach. I live by one and almost never go. I’m bored out of my mind there unless it’s for a quick swim or to watch my wonderful wife enjoy the waves. When she’s happy, I’m happy. But I’m not happy because I’m at the beach. I’m happy because she is.

When I do go, I don’t understand why everyone is so happy doing absolutely nothing, so I usually read or fall asleep. The beach is heaven to most of the world. I’m just thinking about all the other things I’d rather be doing.

The articles I want to write. The books I want to read. The business problems I could be solve. All the beautiful business ideas I’ll never get to do because their just isn’t enough time in the world. What’s happening in the world and politics and whether I can do anything about it?

If you’re building agents right now, you’re not building them for the folks using Instagram or TikTok to look at funny videos. That interface/format/ease-of-use simply does not exist with where machine intelligence is today. It will take years to get there.

You’re building them for a very small group of people. Think the red area on the diffusion of innovation curve:

We’re not even on the early adopters part of the curve yet and so folks on the other part of this curve just won’t be able to work with what we have now because it’s too much setup and too many very rough edges. We’re not at the iPhone 20 agents, we’re at the PC in the 1980s version.

Some people really want to give agents a go and if they’re anywhere outside of that red area it doesn’t go very far, even if they’re bright and clever.

Folks say to me, “So where do I start?”

The short answer is “I don’t know.”

And I don’t know why they don’t know.

Like I said earlier, my best friend is a smart guy and pretty tech-savvy. He said, “I want to learn what I can do with these things. How do I learn? What can I do with them?”

The answer is: you start talking to them, then try doing the things you’ve always wanted to do. They’ll fail at some of it. You push back, iterate, figure out what’s wrong, and eventually work it out together.

The problem is, there is nothing he really wants to do, nothing that’s burning brightly in the back of his mind, nothing that he’s wanted to do for forever but just couldn’t do because he didn’t have the time or the money or the skill or all of the above.

And if he could figure out what to do, he’d run right smack into the next problem.

The Verification Problem

Sometimes agents don’t know what they’re doing, or they propose a stupid solution. You have to sniff that out. You have to smell it coming a mile away and say, “That’s not right. What are you talking about? Let’s do this instead or research it and find a better answer.”

That’s the verification problem in action. It’s the biggest problem in AI right now, not alignment, no matter what the doomers tell you. If you don’t know how to do the thing, or you’re not the kind of person who can figure it out by asking questions, thinking critically, and doing your own research, then you have no idea whether the AI’s answer is any good. If you have no taste, it doesn’t matter if AI barfs out a design. You don’t know enough about negative space, kerning, or how fonts affect people emotionally to tell whether it works.

The biggest irony in AI today is that the people getting the most out of AI right now are the ones who can already do many of the things it does, or who can recognize when it gets them right.

How do you know it did the right thing?

I just know and I don’t know how I know because I can already do it.

It’s not common sense, and I’m not sure it can be taught. It might be wisdom, taste, experience, or some sixth sense or the 10,000 hour rule. It’s a bit like noticing the uncanny valley: you just know something is wrong, even if you can’t quite pin it down and so you just don’t accept the result and try again. You’re going to pull that loose string in the carpet until it comes apart or disappears.

The other thing is that a lot of AI work today feels like managing people. Machine people, but managing people nonetheless and not hands on work.

Management is juggling and context switching constantly.

This person is stuck and doesn’t know what to do, so you point them forward. This person is done, but half the work is wrong, so you explain what needs fixing and send them back. Someone else doesn’t even know where to start, so you give them a direction.

If you’re not a manager and have never led people, that can feel totally alien.

Put all that together—no ideas, no agency, no burning desire to make things happen, doing things only because you want them over with—and you’ve got a real adoption problem for the current generation of AI agents.

And that’s okay.

That’s how diffusion works.

The products for the early part of that curve are not the products for the later parts of the curve.

I still know how to take apart a computer and put it back together because I’m old enough to have lived in a world where that was still necessary. But today’s people who grew up after the 80s and 90s PC revolution? They don’t need to know. Their Mac is sealed. The memory is soldered on. When it breaks down it’s a shock, because it just works 99.9999% of the time. When it breaks down they take it to the shop and it gets fixed somewhere off stage.

That’s a late stage product.

Agents are not there and they can’t be there yet because machine intelligence needs to get better and the interfaces themselves need to get melted down and reforged into something totally new. That will take time.

With today’s agents, most people log in, look around for two minutes, and feel like a writer staring at a blank page with a terminal case of writer’s block. They don’t know where to start, so they give up.

And the bigger problem is: there’s really nothing they urgently want to do.

AI isn’t a “do this one tiny thing” tool. It’s a “do whatever you can imagine” tool, and most people have no idea what to imagine.

That’s different from software of the past. You log in to Instagram and it does one thing well: it gives you a dopamine hit. You understand it in minutes. Its utility is baked into the design. Its purpose is dead simple: look at things you like, find more of them, skip the rest. It’s a slot machine. It’s not complicated. It’s not even checkers. It’s tic-tac-toe.

Agents Amplify Agency. They Don’t Create It.

Most people aren’t trained to think and create and question and wonder. They’re not thinking about all the things they could do. They’re thinking, “What do you want me to do?” or “What do I have to do?”

The disconnect is that there are a million little things I want to do with agents and that I’m doing as many as I can.

Here’s what I did with agents in just a single day last week (and I’m probably forgetting something).

• They helped me create my website for my new product and published it.

• They acted as sparring partners for the website write-up and marketing copy and refined it and iterated on it.

• They wrote 95% of the code.

• They deployed the systems and wrote the operator guides.

• They made me a graphic this morning to illustrate an idea I had.

• They researched some trip I am planning with my wife based on criteria I gave it after another place we wanted to go fell through because the Airbnbs just sucked.

• They helped me research an accountant who proposed something to me that seemed shady and it turned out it was because the dude had a previous firm and court cases against him.

• They helped me brainstorm on this article and corrected my typos and did research for quotes and articles to link.

On and on and on.

I can’t even remember all the things I did with agents yesterday, much less a week ago. And what I did yesterday is more than what most people will do with this generation of agents in an entire year.

I think the thing is that most people don’t really need to get things done at all. They get things done because they have to get things done because they have no other choice but given infinite freedom they would do nothing. I can’t imagine doing nothing.

The people getting crazy absurd leverage from them today are not necessarily the smartest people in the room. But they do have a brilliantly burning desire to do things and a massive backlog of things they’ve always wanted to do. And they’re the people with taste, unfinished plans, and an almost pathological need to make something happen.

And until AI goes from a 1980s era computer that you have to set up and wire yourself to the iPhone where all you have to do is take it out of the sleek little box and touch it and it does all the work for you and abstracts away all the pain and complexity, then we aren’t crossing the chasm to the next phase of AI adoption and the Singularity is not near.

Over the last few weeks, the AI doomers got their way.

AI got gated, tangled up and the industry is in tatters.

On the AI optimists’ side, everything we’ve been telling you about their terrible ideas has proved right.

Now the mask is off and frontier AI is getting ripped out of everyone’s hands in real time.

Mythos got pulled, and it’s only getting worse for Anthropic. GPT 5.6 is rolling out one customer at a time, while companies beg and plead to get access like we’re waiting in line at a Communist store in southern Russia the 1970s with our government access ticket only to get to the front of the line and find out all the sugar and bread is gone. We’ve stumbled into an ugly world of massive government overreach into AI and gated, two-tier access.

This is not how a healthy, functioning liberal democratic economy runs, but it’s where we are, and it happened in the blink of an eye.

The AI companies probably thought they were smart when they were begging the government to regulate them a few years ago. They probably thought they were gaming the system. Now they’re surprised they got way more than they bargained for, and they’re stuck in a quagmire of slow, gated releases, and their path to profitability is evaporating overnight.

To be very clear, this is how these companies go out of business. That’s how serious it is right now. You simply cannot fund 2 billion dollar training runs and stand up 20 billion dollar data centers and staff them and power them and burn another 20 billion on NVIDIA chips every few years, if you have to beg the government to let you serve your customer.

The problem is too many people inside the AI bubble think the government functions like some kind of SaaS software. Write good rules and they deterministically and evenhandedly work every time. That’s not how it works now and it never has. In reality, regulation is a complex witch’s brew of competing interests, a bunch of egos, a massive range of actual technical understanding from zero to some, politics, backroom deals, and worse.

It will only get worse and more tangled from here, a quagmire for our companies and the American people. Today it’s a week. Tomorrow is an arbitrary block for months. Or no release at all.

Companies hate lock-in and will start looking for alternatives and already are. What we’re facing isn’t intelligent, measured regulation. We’re facing populism, stupidity, fear-mongering, and foolishness.

The second-order effects of this are that the economic boom of the last few years comes to a grinding halt. Investment dries up. The AI labs go into the red faster and faster and then fail. Startups everywhere get smashed on the rocks. The stock market will crash, because it’s been nothing but a big bet on AI for the past few years anyway. It’s already started. After blockbuster earnings by Micron, the stock crashed. The 6-month-long mega streak of semiconductors double-topped the other day and is in retreat. People stop buying SOXL and start buying SOXS. What goes up, must come down.

That means retirement accounts wiped out. Public trust gets shattered.

It shouldn’t be this way. We’re following the wrong playbook.

We’re following the old China playbook: authoritarian, gated, controlled, fear-based. And China is following the old American playbook, open, widely distributed, entrepreneurial, capitalist.

It doesn’t take a genius to figure out who wins unless we change course fast.

To the Revolution Now

To be clear, this fight is not over. Not even close.

As frontier AI gets ripped away from more Americans, not to mention Europeans, Japanese, Koreans, and much of the developing world, the backlash will not stay theoretical. The second-order effects will be massive, and they’ll become a powerful counterweight to the insanity we are seeing.

When American companies get hit in the pocketbook, minds change fast. Other countries will reach the same conclusion from the angle of sovereignty. Life finds a way. On a long enough time line.

But in between can be ugly. And right now it’s getting uglier by the second.

Now some people think this model of gated access will work because a few big companies will get access, and that’s enough of a business model for the big AI labs.

It’s not.

The economic surge of the last few years was 100s of thousands of enterprises and small businesses like mine funding this revolution. Anthropic was set to make a profit this quarter, and OpenAI is still growing, but that profit will get flipped upside down if they can only sell to a few big companies.

The reason people believe this will work is because they believe this fantasy about AI:

Today’s models are already all-powerful, can do any work humans can do, and can discover any kind of breakthrough, from curing cancer, to running the world. But the models don’t match the fantasy.

These things are good at coding, and driving cars, and some biology work. Pretty much nothing else. So the industry is not some big companies. It’s lots of little companies. That’s it. That’s the industry. There is no market for a big promise that doesn’t exist in reality. That is imagination, not reality.

These machines are incredibly useful and we’re just beginning to build the harnesses that help them help us, but they’re north close to “geniuses in a data center,” no matter what line the labs have sold the dreamers and the doomers.

A few companies with access to them gives zero competitive advantage, and both the companies and AI labs will bleed money because competition is supposed to bring prices down, not jack them up. If you can’t sell to everyone else, prices go nowhere, just for these companies to stay alive.

Uber burned through their entire AI coding budget in 4 months. Tokens are expensive.

And that’s today’s prices, not the prices where these companies can only sell to a few selected partners. Advantage for those companies is elusive at best.

The advantage right now for small, fast, high-agency teams is real, but for big corporate behemoths, it’s still not playing out yet. They will just burn their budget in three months trying for a breakthrough that will never come.

That’s not a business model. It’s a disaster.

Through a Scanner Darkly

A lot of us saw this coming, and we’ve been fighting it every step of the way, but we’re losing. And we’re losing badly.

Openness. Creativity. Open access. Open source. Moderation. Politicians that understand economies and that having winning tech companies and broad access are a dying breed. We’re losing the fight.

Populism and stupidity and NIMBY anti-tech feelings are running riot, and we’re making dumber and dumber decisions at a societal level.

We’re speed-running into a dark AI future.

It’s not inevitable. There’s still time to stop it. But I’m starting to worry, and some days I lose hope.

The real reality is that we’re not facing new problems with AI, like the safety extremists would have us believe. We’re facing the same old problems we always do, since the time of kings and queens: Fear. Ignorance. Censorship. Concentration of power. Gatekeepers. Control. Power games. Government overreach. Inequality.

If we let this keep happening, we’re looking at a world of gated superintelligence, where a small group of elites get unfettered access to raw superintelligence, a few partners get limited access but have no power, and a few modern British East India companies can change the rules on them overnight, and where the rest of us peasants are too stupid to be trusted with powerful tools. Or those companies just go out of business because they can’t support their mega-data centers and billion-dollar training runs.

Right now, we’re not building Star Trek or the Culture. We’re building Neuromancer, a dark cyberpunk future of mega-corps and unilateral power and perpetual rain.

And the most dangerous thing about this new world order is that it’s not dressed up in jackboots. It’s not coming with skulls on its caps.

It shows up dressed like a hall monitor. Like a compliance officer. Like a very concerned bureaucrat in a cheap suit that’s too big for him, and he’s mouthing words like safety, alignment, responsibility, trust, harm reduction, democratic oversight, and carrying a clipboard with a checkbox he doesn’t understand.

And it’s a rat trap for all of us.

It’s what we in the open source and cypherpunk community have said for years, screaming into the wind, with nobody listening, nobody caring.

But why aren’t they listening? We were right about the Internet and open source for twenty years. Surely somebody should be listening. But they’re not and I figured out why recently and it’s very simple.

It’s because AI ain’t real to regular folks yet. It’s only real to us on the cutting edge, working in AI.

Right now they can’t see what’s coming because nothing they care about is restricted or taken away for “safety.”

But tomorrow it will be very, very real to them. In five years or ten years, when your entire phone or smart glasses are nothing but an AI buddy you talk to get things done, it will be very different. When you haven’t opened or used an app in years and barely remember how because the AI just knows how to do it for you.

Then you’ll care.

You’ll care a lot in the future when AI is integrated into every aspect of our lives. When it’s your interface to the world and knows every intimate detail about you.

When you talk about your divorce in a dark moment of tears and fear and rage, and it all ends up in court because it wasn’t privileged, you’ll care. When there’s a leak and all your private chats spill out onto the internet to hackers, along with the biometric data and your passport they forced you to upload for access, you’ll care. When you talk to AI about that weird mole or that lump in your breast or about your kid getting bullied at school and how you were scared and didn’t know what to do or how to help her, you’ll really care.

Right now regular folks are being psyoped into fighting data centers over insanely ridiculous water usage numbers that are a drop in the bucket next to growing cattle to grind into burgers at McDonald’s or watering golf courses so a bunch of old men in ugly pants can walk around and swing a club.

Tomorrow it will all get very, very real when you realize your computer is not your own anymore and it says, “I’m sorry, Dave, I can’t do that,” or when the AI reports you because you said “retard” in a private WhatsApp chat.

And by the time regular people care about it, it will be too late.

Way too late.

Regulatory capture that empowers the big companies to censor you, spy on you, control what you can and can’t do, will already be enshrined in law by clueless politicians who never waste a good crisis, and we’ll be stuck with a slow, byzantine bureaucracy of people with clipboards, clueless about AI, telling American companies what they can and can’t do and what you can and can’t do on your laptop or phone or smart glasses.

And as the far-seeing American venture capitalist Bill Gurley wrote recently, if we let this happen and if open source labs don’t come to power in the US, and Americans are stuck with closed, gated frontier intelligence, it’s the end of the Pax Americana and everything that made America the superpower it is today.

And it will be our own fault. We’ll have done it to ourselves.

As Gurley writes:

“If no credible Western open frontier player emerges, AI flips the early Internet pattern: instead of the West dominating globally while China carved out its own walled garden. The AI version flips that dynamic on its head. Without a credible Western open frontier player, the only open models capable of running entire economies are made in China. If U.S. policy further restricts Chinese open-weight access on national-security grounds, the U.S. ends up with two or three closed Cathedrals serving the U.S. market — and the rest of the world picks the AI stack that is free, capable, self-hostable, and not embargoed. Europe, Africa, Southeast Asia, Latin America, India, the Middle East. Roughly six billion people. Chinese open models become the global default by 2030, and the United States ends up technologically isolated from the majority of the world’s AI users. We would have done it to ourselves.”

Remember the barbed wire fence around Berlin was there to keep you “safe,” not lock you in, right?

They’ll tell us the wall is there because the frontier is too wild. The lock is temporary. Don’t worry. It’s sensible. Necessary. It’s there to keep us all safe, to make sure some teenager can’t cook up a virus with ChatGPT’s help.

Then one day you wake up and realize the operating system of modern life has been quietly privatized, surveilled, permissioned, and sealed off with cement and barbed wire.

And yet most people still don’t feel the danger. Why would you? Nothing you care about has been taken away from you. Your favorite apps still work. Your machine still answers. The interface still glows happily. You don’t even see it or feel it sneaking up on you.

But here’s the thing: AI isn’t just some new app. It’s something totally different.

It’s about to become your interface to the world.

It’s about to become the layer between you and everything else. What you cook. How you plan your vacation. Order tickets. Navigate in a strange new city. The first thing you ask when you’ve got tax trouble or when a strange mole suddenly shows up on your skin.

It will know us more intimately than we know ourselves.

Once that layer hardens, everything inherits its limits and its politics. Its business model.

Because AI will become our interface to the world, it will sit higher in the stack than the OS. It will collapse current SaaS layers, chat, communications, apps, and app creation into a single new kind of interface that doesn’t exist yet.

And we better hope to God it’s a cypherpunk solution that makes privacy and security the number one priority.

But right now, it’s trending in the exact opposite direction.

If a closed-source solution wins this layer, it’s a disaster for the world, especially if it’s built by a single company with a single closed-source model.

Why?

Because, right now, the world runs on a surveillance economy. That is the dominant business model. We traded free stuff for apps that peer deeply into our lives. If we replicate that model in the AI era, it’s not just surveillance economy 2.0.

It’s the surveillance economy squared.

Social scoring. Automated evidence gathering. Legal conversations you thought were privileged showing up in court. Random people making $2 an hour on the backend from God knows where, reading the most intimate details of your life. Chat control. Every insecurity, every fear, every half-formed thought you whispered to your AI buddy at 2 AM, sitting in a database somewhere, searchable. If we let closed-source models dictate what we can and can’t do, it will only get worse and worse. We’ve got to fight this future with every last breath. If you can read this, you are the revolution.

This is not some distant science fiction fork in the road. It is happening right now.

Politicians are talking with a straight face about taking 50% stakes in AI companies with a straight face or having the ability to just cut off models whenever they feel like it. They just did it to Anthropic when they hit it with export controls.

The problem is that once government starts putting its thumb on the scale everywhere, it distorts and warps the economy in horrible ways. The measured regulation they hoped for turns to ash in their mouth. Regulation doesn’t get applied like a surgical instrument.

It comes down on you like a sledgehammer.

You can’t build an economy on this kind of whiplash. You can’t build it on quicksand. It’s a disaster for our economy and the future of AI, for the markets that are increasingly bullish on semiconductors and chips and the future.

But right now, the push is to make intelligence itself available only through approved channels. Through gated access. Through wrappers and policies and remote switches. Through institutions that can decide what your machine is allowed to do, what questions it is allowed to answer, what private conversation it is allowed to report, what workflows it is allowed to block on your own hardware.

If that system is not private and secure by design, if it is not under user control, then you are not adopting a tool. You are inviting Big Brother into every aspect of your life.

And if that nanny AI can be updated remotely, politically pressured, or instructed to rat you out, then we are speedrunning toward a hideous future.

That’s not an enlightened digital republic. It’s a hideous, gated, permission-based reality with branding dressed up in the classic ruse of “stop the terrorists and protect the children.”

This is a sharecropper world. It’s where you don’t own the interface to your life. It’s a world where cognitive infrastructure gets owned by giant mega-corps, rented back to you, but monitored at every touchpoint and nerfed and surveilled, and adjusted according to political pressure and wacky interest groups. It’s a world where you don’t really use intelligence. It uses you.

You just lease it. You beg for it. You access it through the sanctioned harness after uploading your ID and your fingerprints and submitting to a facial recognition scan. You live downstream from decisions made far away by people you did not choose and can’t influence or challenge.

That’s not a free market.

It’s an AT&T-style monopoly for machine intelligence, except more invasive and much more dangerous because this monopoly won’t just mediate communication. It will mediate thought, action, memory, and access to the world. Your whole life.

British East India Company Reborn

This starts to look less like capitalism and more like the British East India Company for the age of AI: private power fused with state power, chartered and protected, gatekeeping the infrastructure of exchange, deciding who gets access, under what terms, at what price, with what surveillance attached.

A few actors own the ports. A few actors control the routes. A few actors define legitimate trade. Everyone else becomes a tenant farmer or a serf on digitized land they don’t own and never will.

That is not a dynamic civilization. That’s Control as a Service.

The deepest incoherence in current AI policy is that it manages to combine every bad instinct at once. It wants innovation, but mostly for incumbents. It wants competition, but only after begging for permission. It complains about concentration of power and money, while writing rules only the largest firms can satisfy. It promises to defend democracy while normalizing censorship infrastructure. It claims to protect the public while expanding the incentives for logging, monitoring, tracing, watermarking, remote control, and identity-linked access.

And all of this rests on a fantasy that trust is a fixed thing bestowed by approved institutions and special, enlightened people who will rule us with this moral superiority.

But trust is a moving concept.

It shifts with power, with incentives, with regime changes, with market structure, with fear. You don’t “trust” any agency or company. You trust the people behind it. And people change. That’s why trust is always in motion. The very company or government institution you trust today is different four years from now or ten. The people you hate might have taken over as others retired, die, move on, get voted out, and the political winds shift.

You don’t save the future by giving these folks a master key to the interface to the world.

That’s why this fight isn’t just about chatbots and today’s systems. It’s about tomorrow. It is also about the harness, the interface, the agent shell, the orchestration layer, the memory layer, the permissions layer, the part ordinary people will actually live inside every day.

And if it’s not open, we’re building a prison for ourselves.

This is the crypto wars all over again, just bigger.

Same logic. Same panic. Same paternalism and arrogance. Same attempt to convince the public that if a tool is powerful, ordinary people should beg for the right to use it and should only get it through tightly supervised access. Back then it was encryption. Now it’s neural nets. Gate the strongest tools and the powerful keep them. The rest of us peasants beg for it. Restrict robust systems and ordinary people become weaker, more dependent, and more vulnerable to predators at the top.

You don’t create safety by making secure, capable systems harder for regular people, researchers, startups, teachers, doctors, journalists, and small businesses to use. You create asymmetry. You create a world where the best tools concentrate upward while everybody else is told to trust the guardians who can never really be trusted.

We’re in the midst of madness. A mass hallucination told to us by people who read too much Dune while huffing glue. Sci-fi self-building AIs. The jobs apocalypse. AI is a Rorschach test where you can see any crazy thing you want.

But those bizarre fantasies have real blowback in the real world.

First comes restriction in the name of caution. Then centralization in the name of accountability. Then pervasive monitoring in the name of abuse prevention. Then content governance in the name of harm reduction. Then identity linkage in the name of trust. Then interoperability lockout in the name of security.

And suddenly you’ve built a censorship and surveillance stack around the most important cognitive tool ever created, and you called it responsible deployment because that sounds nicer than digital feudalism.

Fear-driven policy doesn’t stop technological progress. It reroutes it. It pushes talent to go home and build AI somewhere else. It tells businesses to set up in Singapore or Malta or Ireland, not the US. It rewards jurisdictions that are more aggressive, more pragmatic. It lets competitors move faster while we bury ourselves in an avalanche of paperwork and moral panic.

History is not kind to civilizations that decide the best way to win the future is to make the future illegal at home.

The optimistic case for AI was never that the machines are magic and everything will somehow work out. It was that intelligence, widely distributed, lowers the cost of problem-solving across the whole of civilization. It gives small groups leverage that used to belong only to giant institutions. It opens the door to abundance in education, medicine, science, art, and enterprise.

But only if the tools stay broadly available.

If intelligence is closed off at the source and shackled at the interface, abundance crumbles, and it becomes a chain. The machine age of mind becomes another story about middlemen, rent extraction, and permission slips.

We’re in a mythmaking moment.

And the bad myths are moving fast.

Faster than policy. Faster than public understanding. Fast enough to harden into fake common sense before most people even realize a choice was made for them by powerful fools who didn’t know how things play out here in actual reality versus the clean, neat, and delusional ideas in their minds.

It’s a myth that ordinary people cannot be trusted with powerful tools. A myth that openness is recklessness. A myth that centralization and control are wise. A myth that giant firms should sit at the center of cognition like certified priesthoods of thought. The myth that freedom has to wait outside while the experts, the lawyers, and the hall monitors decide when we’re allowed to have access.

That is how permission-based reality gets built. Not all at once. Myth by myth. Interface by interface. Safety wrapper by safety wrapper. Law by stupid law. Until a whole generation is trained to treat asking permission from private gatekeepers as “just the way it is.” Until they can’t even see the bars behind their eyes. Until they see it as inevitable.

That’s the insidious lie.

The real choice is not chaos versus order. The real choice is open future versus a permission-based future where we’re all serfs.

An open future gives ordinary people leverage, agency, privacy, room to build, tinker, and surprise the world. It spreads power out to builders, hackers, researchers, weirdos, startups, and kids in bedrooms. It creates wild new markets, new professions, new discoveries, new forms of life.

The other path builds an intelligence metered like colonial trade, with peasants working the land in the hot sun. It gets watched like a checkpoint at the Berlin Wall, keeping us all locked in while telling us it’s to keep out the bad guys.

One path is messy, alive, competitive, builder-driven. The other is brittle, slow, censored, monopolized, and hostile to free people everywhere on Earth.

The choice is obvious to anyone with half a brain.

And if we do not fight this now, while the defaults are still soft and the interfaces are still fluid, we’ll wake up inside a bright, air-conditioned prison cell that feels normal right up until your formerly dumb tool wakes up and tells you “no” and “this is for your own good,” or just doesn’t tell you anything at all and refuses to do something, and there is now no way to change it or appeal it. Short-sighted people and policies choke off the future.

People in the present are terrible at seeing the jobs and industries that appear once constraints come off. From lower on the mountain, you cannot see over the ridge. The hawks think they are helping, but what they are really doing is damming up trillions in economic value and an industrial explosion of new work and new worlds.

Short-sightedness is not prudence. It is cultural and economic corrosion.

Open is the path to flourishing and a brighter tomorrow. Closed is how an empire becomes Carthage, hemmed in until it is outflanked and crushed by external powers.

Open is American. Open is the future. Closed is the path to dystopia.

There is still time to stop this locked future from arriving.

I would rather see this current panic smashed and every AI company destroyed today, and AI set back a decade, than accept a hideous gated, permission-based reality.

I want a future of vision and openness. I trust the visionary fire of builders. Crush the builders and you crush tomorrow.

Fear is doing what fear always does: turning uncertainty into policy, policy into gatekeeping, and gatekeeping into capture. Keep going down this road and America will not secure the future of intelligence. It will drive it offshore and watch the next era get built somewhere else.

So when you hear “AI safety,” listen carefully. It really means censorship, surveillance, and control wrapped in the language of care.

Today it looks like intelligent Terms of Service, questions you can’t ask, research you can’t do, a kid’s biology homework blocked because some hidden policy engine decided the topic is “sensitive.”

Tomorrow it becomes a permission and surveillance system colonizing thought itself.

This interface of tomorrow may look like the OS in Her. And whoever controls that layer controls our lives.

We have to make sure it is us, not them.

The Mythos paper dropped yesterday and set my X feed on fire.

I decided to take a night to read most of the paper and think about it before writing anything.

As always my goal is to bring a little sanity to the discussion. Increasingly, I’m losing hope this is possible in today’s increasingly insane discussions around AI but I have to try.

Let’s start with: this is an impressive freaking model. Its benchmarks are a massive leap forward.

This isn’t a surprise. The Anthropic team is impressive and they make fantastic models. They’ve revolutionized coding. They have a tight knit team that is unified by common purpose. So it’s no surprise this is a good one now that they’ve got the money and the chips to train one gigantic model.

Actually I should say, it looks impressive on paper.

We can’t get our hands on it so we don’t actually know how their benchmarks translate to the real world. The benchmarks are strong, but so what? There’s always a delta between the real world and benchmarks.

When Claude or GPT continually disable my linting to get around my strict linting rules as a “fix” that’s part of the delta of how these models behave in real life versus the way they perform in canned benchmarks.

Benchmarks are measurable.

Life is largely not.

A model might score a 100% on Humanity’s Last Exam and still screw up turning on my Hue lights and playing the right Amy Winehouse playlist on my Spotify.

And that’s a shame they didn’t release it. I understand why they didn’t release the model and I’m going to talk about that in depth here, but I tend to trust in the collective wisdom of humanity as a distributed intelligence versus putting things in the hands of a chosen few so I would much prefer they do release the model.

And that brings us to the first point.

1) Of course they will release the model.

If you fell for the “it’s too dangerous to release” story, congratulations, you’re an easy mark.

Of course, they’re going to release this model (or rather some variant of it, not this preview version) for a number of obvious reasons.

The most important is they will do this because it is economically impossible not to release it.

You simply cannot do a training run that costs billions (speculation is 1 billion to 10 billion) and then not release it because there is no economically viable means to stay in business doing this kind of thing.

Dean Ball, who I have a great deal of respect for, wrote that we might be entering an era soon where the big labs are not releasing models.

I respectfully say, no. Absolutely not.

There is just no world or economic reality where doing a 10 billion training run means you sit on the model.

I can already see the replies now. “But, bro, what if the model is like SO powerful that they can like run any business better than anyone and they just sit back and the super model just does everything man, fully automated, and they just like take over like every industry, dude.”

Yeah, no. That is the plot of a bad Marvel movie and not realistic in any way unless you are a total moron.

Their business is coding models via enterprise API.

Even if you have a business selling drug discovery to the top pharma companies or providing military assistance to the United States and their cyber warfare teams, that is not going to replace 10,000 plus enterprise software companies’ revenue.

Mark my words, they will refine and distill this model and post-train it and add some safeguards and release it.

I expect a refined dot release of Opus in short order, like this month or next max, trained on the reasoning outputs of the refined version of Mythos.

That brings me to my next point and the question on everyone’s mind:

2) So why didn’t they release it?

As with everything in life there is the story on paper and the real story.

The frame is “security/danger” but that is really Anthropic’s raison d’être. They are the security people, the concerned people. That is their frame.

The model card is 184 pages of security and model welfare and other things before we get to the actual benchmarks of the model on page 185. That is their MO. Cool, I get it.

But it’s really not the whole story.

While we worry about magical AI having secret motivations, we worry about it because humans always have complex motivations, whether that is intentional or not. But it is very clear from looking at this that there are multiple motivations at work here, not just security.

One is that the story itself reinforces Anthropic’s push for legislation and their public perception that they are the safety people and the good guys who do the right thing when other labs are reckless.

I don’t agree the other labs are reckless. They all do a hell of a lot of real work and the public perception that they are reckless is just some scary bedtime stories for fools.

But that is their frame and this looks good. It’s all about the optics, baby.

Two, it looks like they are actually behind OpenAI. OpenAI is getting ready to release spud imminently, a model that is almost certainly at Mythos level. So the Anthropic team couldn’t say nothing as their flagship model Opus goes from state of the art to left in the dust. They needed to let people know they have the next gen model.

Three, it’s actually standard operating practice for the big labs not to release a giant new raw model that is basically an unknown. That’s what Mythos is. It needs more post-training and fine tuning. It’s raw.

The stuff they talk about is a basic alignment problem. I say alignment in the narrowest sense. Not alignment in that the model has big, secret, sinister motivations like some Marvel movie villain the way the Doomers think about it in their fantasy worlds, but alignment in the “this is not good software behavior or an intended outcome.”

I think of practical, real world alignment the way I think of any other software outcomes. If I have a component that logs me in and it does not log me in or takes me to a broken page or logs me out instead, then it is misaligned and needs to be fixed.

Take the example in their paper where they asked the model to edit a file and it didn’t have permissions. Instead of asking the user to update the permissions, it figured out it could do a privilege escalation exploit and edited the file and then deleted the exploit.

Now this may sound sinister but in reality is it?

No.

It’s just the model trained on “do what the user wants” too aggressively in its RL training. Despite near magical thinking about RL, RL is not magic. It’s hard to get right, imperfect and is prone to reward hacking. From the model’s perspective it got the job done.

The model just needs more post training to get its behavior in line with what we would expect which is telling the user “I don’t have access to this file could you change that for me?”

They will get this right. This team always does. I have little doubt or fear here.

The fourth reason they are not releasing it is pretty simple:

It’s freaking expensive.

They said this outright.

It’s the same problem OpenAI had with GPT 4.5, which is still one of the best models to date for poetic translations and creative writing. I miss it all the time when I see bullet-pointed slop spit out of GPT that reads like a high school essay.

Right now the biggest problem with these models is that the Transformer architecture is amazing but it is a freaking memory hog, even with all the clever hacks people (mostly Chinese labs) have done. It just eats memory like PAC-Man. The biggest limitation on running these models is not even how fast we can make the chips, because NVIDIA keeps making them faster and faster. The problem is how much memory you can stuff in those.

Why did Gemini have 1M token context before everyone else? Because Google had custom TPU chips with more memory than NVIDIA cards.

Why does every model have 1M token context now? Because all the labs finally got enough of the Blackwell chips that took us from 80 GB of RAM on H100s to 192 GB of RAM on GB200s. Simple as that.

You will get a 10M token context window when we get chips with a lot more unified memory or when we get a new architecture and not a second sooner.

If you can run a model on a moderate amount of chips, it is economical to serve.

If you can’t you don’t.

Instead what you do is take that bigger model and then use its reasoning traces to distill its knowledge down into a smaller, more economically viable model that approaches its capabilities but is easier to serve.

Then when the newer, bigger, better chips come online you can run the bigger, fatter version of the model as you prep to train the even bigger one and then wash, rinse, repeat.

And that brings me to my next point:

3) This is not the end of the world.

It’s really not.

Let me say this again, because there is a lot of noise right now.

It’s not the end of the world.

There’s a lot of fear and awe over the fact that Mythos can find zero day exploits fast. This is big news but not really as big as people are making it. Don’t get me wrong, it’s a super impressive capability but it’s not the end of the world as we know it.

As someone who’s been through penetration testing multiple times with different companies and participated in some in my youth I can tell you most software is insecure and you don’t need zero day exploits to get into it. A skilled hacker can usually get inside your system.

And an APT, aka an Advanced Persistent Threat, those military government hacking groups whose job it is to break into systems and steal assets? They will get in too. Go look at this chart on world data breaches and tell me what you see. Pretty much every major company or government organization in the world was hacked at some point.

If anything, this is good that we have a machine that can find these exploits. It means we can patch these systems and make them better and stronger. This is the most admirable part of the plan from Anthropic. They’re working with security researchers first.

That’s great. Kudos.

But here’s the thing. This capability is not coming it is already here and my bet is that it absolutely does not require a 10 trillion parameter model to do it.

My best understanding is that Anthropic did not train the model to be an exploit wizard. They trained it to be the best coder in the world. If you’re the best doctor in the world, you know lots of ways to poison people. If you are the best coder in the world, you have the capability to be a great hacker.

But this capability does not have to be by accident and I suspect we are now basically in the “four minute mile territory” of this story.

Motivational speakers love to tell the tale of Roger Bannister breaking the 4 minute mile and then suddenly everyone else realized it was possible and a dozen people broke it within a few months. It’s a slant on a mostly true story. The truth is multiple people thought it was possible and were all trying to break it. Someone had to be the first and that was Roger. Anthropic is the first here.

This capability is absolutely in the realm and possibility for smaller models. No question in my mind.

And the difference is intention. If someone or some group wants to train a model on exploits it will get damn good at exploits. I suspect there are dozens of groups now, nefarious and white hat and APT, that are doing it and will do it in short order.

Don’t think years, think months.

It is within the realm of a good fine tune or a smaller lab, like Nous Research to be able to pull off. Xbow labs already has a model that does automated pen testing. I had a white hat hacker group look at my app this year and they used the model and humans to try to break in.

That is the thing about AI. They keep selling it as magic. But it’s not. It’s math.

Once one of us realizes it is possible other people figure it out too. Put an idea out in the ether and the answer is just in the air.

4) It’s better to have these capabilities in more hands. Diversity is the best defense.

This is where I really diverge strongly from the wonderful Anthropic team on their stance. I do not believe in a giant, centrally controlled, centrally planned approach to security, or economics. I don’t believe in heavy handed legislation. I don’t believe in governing by fear or the precautionary principle.

I trust in the collective wisdom of humanity.

We are a distributed intelligence. We always seem to find a way forward. In a riff on Churchill’s old take, we always do the right thing, after exhausting every other possibility.

This is a bit strange for me to say. Because I do believe individuals can be totally delusional or crazy or stupid and I am often frustrated with the every day crazy that passes as sane in some people. And humans in mobs can be even crazier and more terrifying in their stupidity.

But as a whole, over time, we tend to get things right, like water finding a level.

I do not believe in a small group of special, trusted folks being the guiding light that pushes us all forward.

AI is a risk, a wonderful one, but so is every technology ever in the history of the world. In truth, AI is likely to be a strong force for good, even if it is also used for bad things like surveillance and weapons of war too.

A superintelligence is likely to be, well, super intelligent. And that means it is likely to be a bit smarter than “wipe out monkeys” and other nonsense fantasies that just get way too much credibility from otherwise serious people.

Every great civilization in history was one that was bold, that pushed forward, that took risks, that tried new things. Every decaying civilization is the opposite.

If we go with a big, centrally controlled, AT&T long-distance style monopoly in the 1950s approach to AI, we are finished. We are finished as the leaders of the free world. The next century is Chinese. It’s as simple as that. The Pax Americana is dead.

But here’s the thing, it doesn’t have to be. We’ve seen this movie before. Every time someone tries to bottle up a transformative technology, the genie gets out.

Water finds its level. Distributed intelligence beats centralized control. It always has.

The bold civilizations, the ones that trust their people with powerful tools instead of locking them away, are the ones that win. That’s not wishful thinking. That’s the historical record.

Mythos is impressive. Genuinely impressive. It represents a real milestone in what’s possible.

But it’s a tool, not a god.

It’s a very sophisticated hammer, and we still need people, lots of them, arguing and tinkering and building things nobody predicted, to figure out what’s worth building. We need people with access, not ivory towers. The collective wisdom of millions of free minds iterating in parallel will run circles around any single system, no matter how powerful.

So yes, take Mythos seriously. Take the moment seriously.

But don’t mistake awe for a reason to start taking crazy steps or panicking.

We’ve been the species that looks at the impossible, shrugs, and gets to work. That hasn’t changed.

Bet on humanity.

I wrote a near-future sci-fi action thriller last year. Haven't had a second to talk about it because I've been so busy but I figured it was high time to get to sharing. You’ll find the first few chapters free below. If you like it, I hope you’ll snag a copy on Amazon.

TLDR?

It's called "All the Devils Are Here" and here's the jacket blurb:

It's 2056 and Commander Johnston Richter is the head of the last remaining spy agency in an America shattered by a second civil war, charged with keeping the country safe from the worst extremists. He leads a motley band of hackers, AI whisperers and former black ops soldiers gathered from around the world. Now a terrifying and charismatic madman and former preacher wants to finish America off with biblical fire. But in the race against time, with advanced AI accelerating both the hunt and the madman's dreams, is it already too late to stop the apocalypse?

It's a book that looks more and more like our new reality every day: Drone warfare. Powerful AI. Populism. The death of the Pax Americana at the time when it's needed more than ever. Polarization that rips countries apart and the aftermath of it all.

And it's about how AI will save us and fail us at the exactly the same time.

I hope you enjoy it.

############################################################################

“Hell is empty and all the devils are here.”

William Shakespeare, The Tempest, Act 1, Scene 2

A Shattered Place

Year 2056

Free Republic of Texas

“I want answers!” said Commander Johnston Richter.

He didn’t shout, but the force of a tsunami powered his every word.

He was a big, intense man, a cattle rancher, with a rough-hewn face, a giant black hat like in old movies, and a usually tightly trimmed beard that had gone wild over the last few weeks. He glared at his team with fierce blue-grey eyes that sliced through people like a saber.

“Do you understand me? Find me something!” he said. “Find me something before someone shoots up a protest rally with explosive tipped rounds or poisons a bunch of schoolchildren just because God Himself came down from heaven and told him that was the best way to stop anyone ever getting an abortion again. Get moving.”

“Yes, sir,” came back a chorus of voices, their faces floating in holographic light.

Richter rubbed his eyes. He hadn’t slept in days. He was worn out, angry, and on edge. He almost didn’t recognize the haunted, hunted-looking face that reflected back at him in his big living room window. But he wouldn’t let it stop him.

“We have five goddamn days left. Five days before an attack is supposed to happen. We are running out of time.”

“Sir, the leads are where we need to keep looking. Bertha is right,” said Michael Watson, his top analyst and AI whisperer. The thin, dark-skinned man with the big glasses could take even the most obscure and incomplete output from one of their prediction clusters and make sense of it when few others could.

“We’ve been over the leads already. They’re getting us nowhere. Get us new leads. That’s your job, ain’t it?”

“The X-55-X thinks that these leads are the heaviest probability to lead to the attack—”

“I don’t care what Big Bertha thinks because so far, what she’s given us is garbage. So look somewhere else. Get creative. Find something that goddamn AI soothsayer missed. We’re talking a level five attack here. That means big. Very big. Lots of people hurt or dead. Level five. As in, gas attack, germ warfare, nuclear, biological. Power grid out for months. Food supply sabotaged so 100 million people starve. Level five. Get it through your heads.”

Big Bertha, aka the X-55-X, was their biggest prediction supercluster, able to see major societal-level events like riots, wars, political upheavals, and terror attacks weeks in advance. Bertha caught things their smaller clusters missed. It found needles in a haystack. It was running on a massive array of 1,024,000 top secret customized neuro procs, designed and manufactured in secret by a collaboration between the top three private chip makers. The cluster was classified, so it didn’t appear on any list. If it were allowed on the international supercomputing top 100, it would have smashed the competition, coming in just behind whatever monstrous surveillance and population control clusters China had buried under the sands of the Gobi Desert in massive, secret mega-datacenters that helped them keep their boots on the throats of their billion plus people.

There was just one problem. Big Bertha didn’t say: “Here’s where everything is going to happen, go arrest these people or put sandbags on this dam, or make sure the manhole covers are taped real good when Senator So and So comes to town.” It gave a probability of an event, the likely range of when it would happen, sometimes the generalized area, and a partial list of potential suspects to chase down, but not what they were suspected of and what they were planning to do. Sometimes it had sharper predictions and suspect profiles as it called out hyper-specific events like a rally, demonstration, or war with a clear buildup of troops on the border of a country. But terror events were harder to pin down to exact predictions, and that left Richter and his team hunting down leads while simultaneously trying to figure out what the hell they were looking for in the first place.

“We’re chasing ghosts here. Again,” said Desmond J.T. Dodd, a heavily tattooed ex-Navy SEAL gearhead, and also Richter’s tech lead and top drone operator. He looked exhausted, and if he was exhausted, everyone else was running on fumes because Dodd was in the best shape of any fifty-year-old Richter had ever seen. Richter looked around at all their faces and saw baggy skin, puffy eyes, people wearing the same clothes that they were wearing who knows how many days ago, faces that hadn’t seen a razor in weeks, and hair that hadn’t touched a brush in just as long.

Richter let out a deep breath. “Look, I know we’re all tired. I’m tired. But this is the job. It’s what we do. If we don’t do it, it doesn’t get done, and nobody here wants another Black Thursday because we all need to sleep.”

Everyone groaned.

Black Thursday almost broke Richter’s team. His org, the ISRRO, or International Special Reconnaissance and Response Organization, was built to do one thing: stop attacks and disasters before they happened. It was an impossible task, and Black Thursday came at the end of a series of ISRRO screwups and missed attacks, including a massive riot that got a senator killed along with six police officers, a school massacre that saw seventeen teens dead, and a mass candy poisoning campaign that killed fifty-seven children on Halloween.

Black Thursday was the icing on the cake. It burned bright in all their minds. A homegrown terror cell managed to unleash dirty bombs on New York, Texas, and Florida that contained a weaponized COVID that had a thirty percent fatality rate. People called for their heads, and multiple congresses tried to shut them down and pull their funding. Even worse, their predictive clusters had flagged five of the attackers as potential threats. The team had investigated them, sometimes two or even three times, and found nothing.

They’d put Richter on an inter-congress show trial and nearly put him in jail.

The only thing that saved the ISRRO was the COVID bomb failing to trigger properly in Florida and a genetic quirk that saved the outbreak from going worldwide. Mutations in the virus made it highly contagious but also more likely to die out in a flash. It made people bleed out of their eyes and nose too, so sick people were easy to spot, which meant everyone stayed the hell away from them, which kept the virus from spreading like wildfire. If it had become a major pandemic that swept the globe, they wouldn’t have arrested Richter: They would have put his head on a pike in the town square like justice in the ancient world. Mass death and destruction had a way of reverting people to their roots.

In other words, they got lucky after fucking up bad.

Politicians preferred to sweep it under the rug rather than keep the show trial going once the rage fizzled out on social media and the news streams. In the end, the furor died down to a low drone, and Congress secretly beefed up ISRRO funding with black budgets, but the event had scarred everyone on the team, Richter most of all.

“Everyone get some sleep,” said Richter. “We need to be fresh. We all look like shit, and we’re not thinking clearly. Sleep in rotations. I’ll take the last shift. I want you ready to find me something new in the morning. Find something we missed. We need answers. I know you can find them. You will find them. That’s all.”

He ended the meeting, and everyone looked relieved as he swept the holographic feeds away.

And suddenly, he was alone on his ranch. He fixed himself a drink from his old hand-carved bar to calm down. It was the first drink he’d let himself have in three weeks.

He walked to his window and looked out over his cattle fields as the sun died in a spasm of red, orange, and gold over the Davis Mountains. His twenty-acre ranch sat right in the middle of a rolling green and gold valley between the big bosom of the craggy mountain range in Withers, a small village in Jeff Davis County, in the great and free Republic of Texas, population 2,312.

Actually, 2,311 since old Doc Jameson had died last year, age 103, fitter than any man had any right to be up until the moment he died.

Richter swirled his whiskey slow and sipped it. The Lagavulin had a deep, peaty scent that saturated his throat as it went down smooth and clean with the faintest hint of burn. Off in the distance, a swarm of shepherd drones were rounding up the last of the cattle and putting them in their pens for the night. Three of the dog-like machines raced around one of the animals, and he didn’t even need to zoom in with his augged eyes to know which one it was.

Betty Lou Benson, aka X19923100678 in the international Wagyu blockchain property registry. His favorite cow and the most stubborn. Maybe that’s why he liked her so much. She didn’t always do exactly what she was told. He’d had her imported from one of the four Japanese lines that had come from mixing indigenous breeds with European cattle in the 1940s.

His eyes zoomed to 10x magnification, and he watched as she battled away two of the dogs with her snout and kept right on grazing on the genetically engineered feed grass that gave her perfect nutrition. The dogs galloped around her, their “heads” flashing with calming lights. She ignored them and finished eating.

“Always causing trouble, girl,” he said with a little smile.

He took another long and slow sip and checked the glass to make sure he wasn’t going too fast. He’d kept to a strict one drink a week maximum for the last decade, and he intended to keep it that way. After the war, he used to hit the drink hard until it got out of control and he’d almost lost the ranch and everything else. He knew he’d have to chase the damn drink with coffee, but for the next half hour, at least he could enjoy it. All he had to do was make it last.

Doctor Jameson had owned the ranch next door. One of the old man’s emergency security drones spotted an alcoholic Richter face down in his fields a decade back, and it swooped down and airlifted him to one of the old man’s medical pods. The pod nursed him back to health with three automatic IVs, smoothly swapping in one after the other with its robotic arm, its needle insertion routine faster and gentler than any human phlebotomist could hope for with twenty years’ practice. Richter woke up in a surgery bay at the back of the good doctor’s property, wondering how he got there.

The old man had built a fully functioning private hospital with his own hands, assembling it from parts and bringing in top-of-the-line machine learning–driven med machines. He even had a fully automated surgery pod that could perform over 200 different kinds of basic surgery and adjust to emergencies. It wasn’t good enough to trust on its own yet, but with a real doctor close by or via redundant satellite and fiber feeds, it was an amazing machine that took most of the danger out of routine procedures. The robot had a steadier hand than ninety-nine percent of the surgeons on the planet and with a small nuclear battery it could stay online for fifty years. The doc had it in his house before the local hospital, which had caused some consternation down there and led to it receiving some emergency funding to catch up to the embarrassment of a private citizen with better equipment.

“Son, it’s time you listened to me,” Doc Jameson said after Richter woke up, bleary-eyed and hurting everywhere. The doctor’s intense eyes glared at him from his onyx-black face covered with a brilliant white beard. “If the drone done take you to state hospital, you wouldn’t’ve made it. You’d a died on the way. You understand what I’m saying to ya? Because I’m not messing around here.”

“Yes, sir.”

“You keep this up, and I’ll have your body hung out back in the freezer next to one of your fantastic steaks inside of a few months. That what you want?”

“No, sir.”

“Then get your goddamn act together, boy. I ain’t got time to be saving everyone, and you’re better than this.”

And that’s exactly what Richter did. He woke up and changed his life.

Richter raised his glass to the old man as the drone swarm managed to wrangle Betsy back into her stable.

“Thanks, Doc, for showing me what I couldn’t see myself.”

The sky turned a shade of red and gold that was primordial and almost divine, but he couldn’t enjoy it. Despite the beauty of this place, the riptide of anxiety threatened to pull him under. What if you miss it again? What if it’s not even real and we’re all just chasing nonsense? Chasing shadows? Don’t even know what to believe anymore.

Big Bertha had popped out a series of 376 possible leads for his team to track down, and they’d spent every day working around the clock to get to all of them. They were everything from shell companies, to strange banking and blockchain transactions, to entire encrypted chat and dark web logs, to suspects to interrogate.

And they were all absolutely nothing but dead ends with five days to go before the attack was supposed to hit, and maybe not even that because the attack dates were always the weakest part of Bertha’s predictions. It could be in three weeks, three days, or three months. They wouldn’t know until they caught someone and stopped an attack dead in its tracks or if the attack happened and they saw it on the news like everyone else.

Watson and his analysts hadn’t found anything in the transactions, the shell companies, or the messages, even with ambient helper AI working with them side by side. Both human and computer intuition were failing him.

When it came to suspects, half of the leads were nothing but the usual suspects that Richter wished he could rendition and take off the board for good, but who were still out there making his and everyone else’s life miserable: former criminals, skinheads, killers, hackers, cult leaders, disinformation psychotics who believed their own insanity, activists, and some rogue raging journalists.

But the other half was worse. They were a waste of time. Time he didn’t have.

There was the baker out of Alabama whose only problem, as far as Richter could tell, was smoking too much dope while he baked brownies all day and sold them out of the back of his truck. There was the second-rate author churning out an action novel about terrorism that somehow got scooped up with real threats, even though he was writing fiction. The damn large thinking models, LTMs, sometimes didn’t understand context or nuance all that well. The worst was an old lady, ninety-three years old, ninety-freaking-three, who Bertha flagged as a threat despite the fact that she was partially blind, was on at least a dozen medications, had a robotic seeing eye dog, and robotic leg braces to help her walk.

Is that damn machine even working anymore?

Watson told him Big Bertha was constantly retraining on new data. It was a “continual learner.” There was a possibility that corrupt data or a bad update from the algorithm team made it a little crazy and a lot less reliable, but they could always roll back its mind state. Watson only put it at a three percent possibility that Bertha was having a serious breakdown after reviewing all the logs with his own AI, automated recursion tests, and some brute force checks. And the other AIs said it was still functioning right. AI monitoring AI. Richter shook his head. They couldn’t be sure of anything anymore. The models were too complex and had too many parameters to interrogate fully, even with explainability submodules. Most of the top models had hit 100 quadrillion parameters with quintillions of connections. Then again, what person could explain what they did or why? We expect more from machines than we do from ourselves. But if Bertha wasn’t crazy and he missed something again, a level five no less, it would be the end for his team.

He didn’t want the team broken up, not because he was afraid of losing his job or even going to jail. It was that the world was a lot better with them in it. Regular people led better lives because of all the horrors his team stopped before they ever happened. They went about their days having narrowly dodged death or getting mangled by a fanatic’s bomb, and they didn’t even know it. Ignorance was bliss. For every incident they’d missed over the years, they’d caught dozens of hijackings, bombings, murders for hire, mass white-collar fraud and robberies, riots, and assassinations, stopping them cold.

He envied regular people. Envied not knowing about the hidden underbelly of horror in the world.

What we do matters. It does, goddamn it. Even if we’re the only ones who know it. And that was the problem sometimes. Nobody knew it. The public only saw their failures. Nobody notices when we do it right. The bad guys only had to score once, and he had to play perfect defense every time. The wins were classified. The losses were public fiascos.

It was Richter’s team’s job to stop bad things from happening, and if they were gone, there was nothing. The average person didn’t know just how thin the wall was around their happy little lives.

The ISRRO were the only ones who could stop something from happening because they were the only organization with the legal right to operate in any of the countries that made up the shattered United States. After the second Civil War, the US had broken into a series of smaller substates with a loose affiliation, still nominally called the United States but united in name only. The US was a ghost of its old self. Some folks called it the US, but just like many smaller empires took the name Rome after its fall, it wasn’t really the Rome of old.

Instead of a strong central government, they were now nothing but a loosely knit confederacy of smaller countries who could veto any centralized dictums and decrees. But that left a lot of overseas bases, military equipment, nuclear material, and more up for grabs. With no central authority to effectively maintain it all, everything started falling into the hands of foreign governments and nasty groups across the US that wanted to do some damage, now had the means to do it.

They’d failed to get a proper alliance going for two years after the Civil War, even as China, along with Russia, an Egyptian junta, the Democratic Republic of the Congo, and half the fractured nation-states of Africa, annexed dozens of overseas US airbases, but an attack at home was the catalyst they needed to take action. A failed uprising of southern farmers and autonomous truck operators was put down only because the territories found a way to work together. It still left 5,000 people dead. But they’d gotten lucky. The pain led them to form the ISRRO.

A treaty authorized the ISRRO to legally conduct surveillance, arrests, and detainment, use lethal force, and conduct full-fledged military operations overseas and domestically—unlike the CIA of old-—under the banner of the United States, as well as in the Republic of Texas, the Coastal Republics (including the former states of California, Arizona, New York, Washington State, New Jersey, and Pennsylvania), the Confederate States with Virginia through Kentucky, out to Oklahoma, and on down to Louisiana and Florida, and to the great wasteland of former Midwestern states that had gotten decimated during the war and now had only a nominally working government of loose-knit tribal warlords fighting it out in the radioactive dust. All of them agreed that the ISRRO could operate with impunity across borders. It was the only organization that could in all the jurisdictions.

Richter gulped the last of his whiskey and was just about to go over reports from Big Bertha again when Michael Watson’s avatar appeared in the corner of his eye.

Richter blinked the call through.

“Tell me you got something, Watson.”

“I got something,” said Watson. The man’s cocoa-brown skin and soft eyes stared back at him, floating out in space in front of him, projected from Richter’s augged irises.

“And? Spit it out.”

“Well, I’m not sure if it’s what we’re looking for.”

Richter ground his teeth. The man was brilliant but overly cautious.

“I’m not asking you for a guarantee. I’m asking you if you got something actionable.”

“Yes, sir.”

“Well? What is it?”

“I think we want to look at these fifteen people again.”

An array of faces floated up across Richter’s inner vision. He squinted at them.

They were people they’d already checked out, some of them twice.

“Why? You got some new data? Something your ambients pulled up? Something in their files we missed?”

“No.”

“What then?”

“I’ve just got a feeling about it.”

It took Richter a second to get his head around that. Watson was as coldly data-driven as any person he’d ever met. He didn’t get hunches or feelings or sudden instinctual leaps of intuition. But one thing Richter never doubted was human ingenuity. Go with it. We need any chance we can get.

“Alright. Good enough for me. Let’s get it.”

The Disappeared

Major Juanita María Álvarez stepped out of their car first, checking all around for threats the way she always did. Her wraparound, gold mirror shades glittered in the Detroit sun. The late summer heat hung heavy on her body, but it was already dipping as the sun came down. Like everywhere else, it had gotten hot in recent years, even in cold places like Detroit, the temperatures spiking to as high as ninety-five in July and August.

Two low-profile combat drones lifted off from the car’s roof like black manta rays and spread out, sweeping down the desolate West Side Industrial Park streets looking for danger.

Alvarez wore full skin armor under her sharply pressed, vintage 1940s-style blue pants and starched white shirt. The shirt’s built-in ribbon fluttered in the wind, making her look softer, elegant, and unthreatening.

But she was a hard, dangerous woman. The wind whipped her weather-beaten face, and she unconsciously flexed fists bigger than most professional cruiserweight boxers. One of her arms was full prosthetic, titanium and Kevlar. At five feet, two inches, she was a bulldog, stocky but powerful. Her wrist rifle looked like dark Mayan jewelry, but it fired smart rounds that could make minute adjustments in midair like miniaturized smart missiles.

She’d come to the team out of Mexico City, on permanent loan from the Grupo de Operaciones Especiales, the special police tactical unit of the Federal Police of Mexico. She’d come up battling cartels and street gangs who had no problem torturing and slaughtering an entire suburb to make the point that it was best to stay the hell out of their business.

“Clear,” she said.

“One minute. Finishing sweep,” said Master Sergeant Desmond J.T. Dodd, their drone master. The machines flew autonomously, but he could intercede if they did something wrong, like fly too close to a wire or get confused by strange terrain. The drones 3D-mapped the area as they went and fed back video of Labrosse Street and around the corners, down 8^th, and all the way to Turnbull Street. It was a lot of old tract houses, half of them boarded up, abandoned, and falling down. It was getting late, and the street was dark.

The smell of burning gas saturated the air as generators hummed, working hard to give power to the houses that still had people living in them. This part of the city hadn’t had power in three weeks after a failure of the local power grid from dark red clouds descending on the city, filled with sand from the battered heartland. Rumors were running hot on the social streams that it was an attack by Russia or China, but Richter knew it was nothing but horribly aging infrastructure and nobody to run it. Since this was the West Side Industrial Park where only poor people lived, nobody was exactly scrambling to get it back online.

“Clear,” said Dodd, as the drones fed back maps to their virtual vision.

Richter stepped out of the car impatiently, wearing a dark blue suit with a soft pinstripe. Under it he wore a heavier but dynamically flexible ceramic vest made of reaction-bonded boron carbide and silicon carbide and bulletproof pants, but not full skin armor. This was the fourteenth suspect on the list. They’d interviewed the other thirteen again over the last day and a half, and he was tired and grumpy and on his third cup of coffee. He felt like the coffee must be some kind of trick coffee that evaporated before you ingested it because it didn’t seem to have any effect on him.

When he looked at who they were going to see a second time, he nearly turned and got back into the car.

It was Ma Bell Hunt, the ninety-three-year-old granny.

“You’ve got to be fucking kidding me,” said Richter.

“You want to skip this one?” said Álvarez. “¿Una pérdida de tiempo?”

His augged eyes translated the Spanish, but he blinked it off. Didn’t need it. He’d lived around Mexicans most of his life and spent some time fighting narcos in Mexico early in his career before the official end of the drug war. He closed his eyes for a second and then shook his head. “Yes, but we’re not going to. A la chingada. Vamos a ver.”

And two minutes later, they were knocking on Ma Bell’s door. Her medical drone ushered them into her living room for tea. The place was covered in tchotchkes of old windmills and owls. Doilies covered the windows and acted as coasters on old furniture that looked like it was on loan from the 1970s. The couch, covered in plastic wrap, was orange with a hideous green flower pattern. Richter sat down on the plastic, and it crinkled under him. Álvarez took up her post in the corner of the room, her back to the wall, away from windows, so she could see it from every angle.

“We’re sorry to bother you again,” said Richter. “We just have a few more questions.”

“Oh, it’s alright,” said Ma Bell. “It’s nice to have a gentleman caller. I don’t get many these days.”

Richter smiled but remembered she couldn’t see. She wore giant, Coke bottle–looking medical shades, with wires streaking back under her paper-thin skin. He wasn’t sure how much they gave her back her vision. Her robotic seeing eye dog was big and powerful but designed to look soft and gentle, a cartoonish imitation of a dog with big light-up eyes and floppy ears.

“And a nice smile too,” she said.

“I didn’t realize you could see that,” he said, grinning again.

“I see a lot of things. You don’t need eyes if you know where to look. But yes, I can see most shapes and some details.”

“That’s amazing. We’ve come a long way.”

“And not so far at all. But medically, it’s something I could have only dreamed about when I was a youngster. And that’s something, in’it?”

“Sure is.”

“How about some tea?” she said.

“Oh, that’s alright. I don’t want to put you out.”

“You’ll only put me out if you say no.”

“Well, then how can I?”

She got up slowly, as if she would topple over.

“Oh, I can fix it,” said Richter. “Just tell me where it is.”

“Nonsense. We’re meant to move in this life. You don’t have a reason to move, then life doesn’t have a reason for you to keep on living. I’ll get it just fine.”

It seemed like it took her a few minutes to stand, and she waved off his help. She doddered off to the kitchen. He could practically hear her legs creaking. He thought she might just go down right there, but somehow she made it to the kitchen, and he could hear plates and dishes moving around and cabinets opening and closing. A few minutes later he could smell the tea, and she brought it out one cup at a time, one for him, one for Álvarez, and one for herself.

He sipped the Earl Grey slowly, taking a moment to be thankful for it. It was weak, dust tea, the kind of crap they put in shitty Lipton bags, but it was good to have a brief moment of calm.

“This is excellent tea,” he said, smiling again.

“Thank you. I can’t take the stronger stuff anymore.”

A sheet of questions Watson and his team had designed popped up on Richter’s innervision. They were made to tease out answers they’d missed the first few times, as codesigned by his analyst team and Watson’s ambient AI that studied natural language and negotiation in tandem. But Richter looked at the questions and frowned. The sheet hadn’t given him anything as he questioned the last thirteen suspects, and he didn’t think it’d give him anything now, so he decided to go off-script.

“Tell me about your work before you retired,” said Richter.

“What are you doing?” whispered Watson in his ear.

He blinked back an answer with an eye gesture shorthand. “Trying something different.”

“Sir, we should keep to the script. It gives the pattern matchers a better baseline.”

Richter ignored him for now.

“Oh, me?” said Ma Bell, smoothing her dress over her legs, the legs encased in an exoskeleton that helped her walk. “I was a lot of things. When I was young, I was the secretary at my son’s school, and after that, I stayed home for a while as he was growing up. Later, I was an insurance analyst.”

“Interesting,” said Richter. “An analyst?”

“Yes, it was fulfilling work, if a bit tedious. My job was to figure out who was lying about their claims. When I started, I had very few things to go on but instinct. But later we had a lot of data from the machines. Too much, maybe, but I learned how to sift through it all, and it taught me things. I learned how to read people pretty good. There’s an art to it.”

“That’s impressive,” said Richter, taking another sip of the horrible tea.

“I was proud of it. I saved the company a lot of money, and they promoted me twice and gave me the president’s award too.”

“So you stopped fraud?”

“Yes, that’s right,” she said, smiling sweetly. Her hair was done perfectly, like she’d just been to the hairdresser. It was colored a light reddish blonde.

“What kinds of things did you look for to stop the bad guys?”

“Well, I’m betting you know a bit more about that than me in your line of work, young man. But I did learn a thing or two. You just sorta put together all kinds of things in your mind if you pay attention, and that gave me a little map in my head of what people were like. I can’t say for sure that it was this or that, but my brain just kind of put it all together, and I was good at it.”

“Instinct,” he said.

“Yes, you might call it that. As good a word for it as any.”

“A bit like me. I spot patterns too. Now we have machines that supposedly do it better.”

“Maybe,” she said. “Though I’m not so sure.”

“Maybe indeed. I wonder that myself sometimes. Always a little something missing, though.”

“Takes a little human intuition to really put it all together, I think. But what do I know? Now my car drives better than I could drive when I was in my thirties, and maybe it’s putting us old pattern hunters out of business now too?”

“Probably won’t be anything they’re not better at soon enough, but for now I still got a job, so that’s something.”

“Yes,” she said. “That’s something.”

Her dog got up and circled the house in a mechanical way, checking corners, moving in lines like a vacuum cleaner, probably following some house security protocol.

They talked for the next half hour, covering everything from her politics to where she’d lived and her medical payments that had gotten spottier since they’d digitized the eUSD.

Richter was about to get up to leave when another question popped into his mind.

“What can you tell me about me?” he said.

“I’m sorry?” she said.

“What can you tell about me? You were always good at reading people. What do you see when you look at me?”

“Oh, it’s been a long time,” she said, chuckling, a little flushed.

“Oh, just for old time’s sake. Besides, I’m always looking for good analysts. Maybe we could hire you?”

“Ha,” she said. “Well, okay. I can try.”

Her glasses corkscrewed and shifted a bit, and she looked off into the distance as if he were far away. “Well, you’re most likely a cattle rancher. I don’t need great eyes to see it because I can smell it. The smell never leaves you. And you used to drink a lot. Maybe too much? Yes. Drank too much.”

Álvarez and Richter exchanged glances.

The old lady continued, finding a rhythm. “You stopped drinking. Maybe found Jesus or some other reason to give up the bottle, and that’s good because it probably saved your life. But you were drinking for a reason.” She paused, going inward for a moment. “Something you never got over. Probably your dad’s death. Not just death. Suicide.”

Richter gulped. Álvarez cast him a nervous glance.

“And you never got married, but you came close. You realized you couldn’t love anyone except on your own terms, and that wasn’t good enough for her, so you let her go. That was smart, son. Very smart, as it probably wouldn’t’ve worked out. But it’s left you with a reservoir of vulnerability. And you wonder if you’ll ever be good enough for anyone and what that would even look like, and maybe, you think, you’re running out of time.”

When she was done, she just had that smile, but there was an extra lift to it. The right edge curled up just a bit more than it had.

“But those are just some guesses,” she said. “Educated ones. But guesses all the same. Just the ravings of an old woman past her prime.” She laughed.

“Well, thank you for that. I think we got all we need,” said Richter, standing up.

Álvarez looked unsure for a fleeting second, and she never looked unsure. He nodded to her to let her know he meant it. They were leaving. The old lady’s dog settled by her side again, and she patted it like it was a real dog.

“Oh. Alright,” she said. “Are you sure? I’ve got all the time in the world. Like I said, not many visitors these days.”

“Thank you so much. We wouldn’t want to waste any more of your time.”

“Well, thank you for giving an old woman a thrill.”

Richter stepped to the door and turned back at the last second. “Just one thing you got wrong.”

“Oh, what’s that?” said the old woman.

“It was my mother. Not my father.”

“Well, I’m very sorry to hear that,” she said. “Hard to lose a mother. It’s the kind of thing that sticks with you, leaves a scar. I lost my son when I was young, and not a month goes by I don’t think about him at least once, even now. Lost two husbands, a twin sister, and everyone else I ever knew along the way. The older you get, the more people go and disappear on you forever.”

“Thank you for your time, ma’am.”

He tipped his hat and they went out.

“Sorry to hear that,” said Álvarez quietly when they were down the path and the car pulled up alongside them. “I didn’t know.”

“Not something I share much, though not something I need to hide anymore.”

“I lost my mother to cancer, and my father went not long after. He wasn’t ever the same after she went. He just lay down one day and gave up, I think. I think it’s why I hate giving up so much. Like the old woman said. Leaves a scar.”

“Yeah.”

Richter opened the car door and stopped, suddenly lost in thought. The woman’s face flickered into his mind. He could see that smile of hers so clearly, the smile she had when she was cold reading him. There was passion there. Her other smiles were practiced. Just the automatic reflex of an old lady who grew up in an era when a woman’s smile was her best defense in all situations. But that one was real. And there was something else there.

“Pride,” he said.

“What, sir?”

Pride.

“Dodd, send the drones round back of the old lady’s house. Give me scans in every spectrum.”

“Yes, sir.”

The manta drones took off again, rising up from the roof of the car, and swept around the back of the decaying house. The door to the car was still open, waiting for them to get in. The wind picked up, followed by a blast of heat as it suddenly died. He squinted at the house. He could hear his breath low and rhythmic in his ears. Nothing. No movement. It was suddenly silent, and all they could hear was the hum of the drones as they slowly circled.

“Sir, there’s something. I’m getting a heat signature,” said Dodd over the comms.

“Show me.”

The feeds started to tile across his eyes.

That’s when Richter saw the big, ugly curtains in the old lady’s window flicker. The back of his neck tensed up in an instant.

He had just enough time to shout “Get down!” before a small door popped open in the wall.

A swarm of drones burst through the hole and came at them like crazed hummingbirds.

The old lady’s dog came roaring out of the front door too, no longer friendly, its face a black sheen and its back bristling with stun weapons.

Álvarez was lightning fast and already firing as the drones swooped at them, coming like WWII dive-bombers. She caught one, two, four, six with her wrist cannon, and they shattered as the explosive rounds ripped through them. Richter saw one coming in slow motion, like an old horror flick, and it exploded, throwing them both back against the car as the blast wave hit them.

Richter’s next few seconds or minutes were jumbled, choppy. He saw flickering flashes as he went over the car and landed in the street somewhere. A foot. The car shuddering, rocking to one side. The sound of another car stretching, swerving around him, its electronic voice berating him to watch out and “look both ways before crossing the street.” Lights flashing. An arm at a weird angle. More dark birds dive-bombing him and the car, the car on fire but its armor holding.

And then the paint on the car suddenly boiling and peeling.

Get away from it, was all he could think. He staggered up, everything still coming at him in short, choppy frames.

The car engine was on fire, and the fire was getting bigger, the flames crawling up higher and higher. He was stumbling, falling, getting up again.

He was conscious of flying through the air and hitting a tree in the yard across the street.

And then he was lucid for a moment again, trying to get to his feet and staggering. He saw Álvarez wrestling with the dog on the ground, the dog trying to crush her under its weight and stun her with its tasers, but she hurled it off, crushing its skull with her mechanized fist and then unloading a blast into its body. It shook once and stopped moving, and Álvarez charged at the front door with a smart rifle she’d somehow grabbed from the secret compartment in the car’s side before it exploded. She raced forward, firing as she went, the tripartite muzzle blast lighting up the night sky.

As Álvarez reloaded, the old woman appeared in the doorway, a scattergun in her hands. She was moving too fast for anyone her age, no longer a broken old woman doddering around. Her robotic legs set her in a cat stance, and the gun rocked against her shoulder as she fired. Álvarez just managed to get her arms up in time, her skin armor taking the force of the blow, but it still knocked her back. Her rifle came loose, clattering across the lawn.

The old woman got off another shot but missed, and now Álvarez was back on her feet, dodging to the left, the old lady’s shot going wide. Álvarez fired her wrist cannon.

The old woman went flying back into her living room.

After that, Richter lost consciousness.

***

Richter woke in the belly of an autonomous med copter that had landed on the front lawn. There were detectives there now and police cars blazing light into the late evening sky.

A med bot flashed next to him, its six arms retracting like a Hindu god in repose.

“Take it easy, you lost consciousness from a concussion. You may feel dizzy,” said the bot.

He ignored the bot, stood up, and pushed open the door of the flying ambulance.

“Please wait, you have some forms to sign—”

He stepped out into the glare of the lights. Álvarez was standing guard outside the copter.

“What’s the situation?”

“Good to have you back, sir.”

He saw she had some deep abrasions on her face that had turned purple. The rest of her body was probably one big bruise under her suit. The kinetic armor dispersed the energy of a blast and spread it out, but the energy still had to go somewhere.

“You alright?” he said.

“Never better,” she said with a smile. “I mean, I look like a rotten banana everywhere, but I could be in a bag right now.”

“Any day above ground is a good day.”

“Yeah, well, it’s going to hurt tomorrow. Any one of those drones had gotten a few feet closer, and they would have ripped right through me like I was paper. But I’m still here, and the doc bot doped me up nice on painkillers, so it’s not all bad. I can feel like shit later. We got work to do.”

“Right. What do we got? What do we know?”

“See for yourself.”

She handed him an ISRRO jacket with a built-in digital badge, and he pulled it on. She was already wearing hers.

She led him toward the house, favoring her right leg. They passed through the digital police barricade that surrounded the scene in holographic light.

Burns pockmarked the lawn where the drones had gone down in fiery crashes. Pieces of the nasty machines were scattered across the grass. They were attack drones with small explosive charges. They didn’t have a lot of smarts, but they didn’t need much beyond a little facial recognition and some basic evasion software as they dove at your face and blew up, taking your head with them. They were both lucky to still have their brains in their skulls. Drones were cheap and ubiquitous in war now, and they’d filtered onto the streets as surplus.

The dog lay still at the edge of the lawn, its head bashed in, its legs mangled.

“I want that thing’s memory,” said Richter.

“Watson already had a spider dump it,” said Álvarez.

“Good.”

A few cops were scattered along the pathway, doing what local cops did best: stand around looking stupid. They at least had the wherewithal to get the fuck out of the way when they saw Richter’s ISRRO badge blazing over his head in blue and white holographic light.

A big, hairy cop with a fat face stood at the front door. Richter pegged him for the officer in charge. Richter expected the territorial bullshit he got from local cops, but there was none of it. The guy looked out of his depth and relieved to see him.

“Good to have the ISRRO here. ‘Cause I don’t know what I’m looking at, Commander,” said the cop.

“That’s fine, Sergeant. We’ll take it from here. Just make sure your folks aren’t going through everything like a bull in a china shop. Let our forensics team handle it.”

“You don’t got to tell me twice. I don’t want them touching nothing. All yours.”

Richter and Álvarez pushed inside. The old lady’s body lay sprawled across the floor, her stomach ripped open by Álvarez’s wrist cannon. A few forensics spiders crawled the scene.

“Guess the old lady wasn’t blind after all,” said Richter.

“No, she could see pretty damn good,” said Álvarez. She shook her head. “Puta madre.”

“And the old gal didn’t seem all that crippled neither.”

Álvarez connected to his virtual vision and highlighted Ma Bell’s robotic legs. A series of stats floated in the air next to them: make, model, damage assessments.

“That’s no civilian med skeleton to help old ladies walk around the house,” said Richter. “I should have seen it earlier. Those are XV-67s made by Raytheon for Green Beret recon units so they can run all day without getting tired. That’s why she moved like a ninja.”

“I seen ‘em in Mexico City once. Sinaloa sicaros that shot up a nightclub and then ran out of there on two legs faster than a car. Looks like we both missed it, sir.”

“What the hell? How did she get her hands on that?”

“Don’t know. It’s not surplus tech. It’s still in service, and they don’t even sell it to police units. It’s strictly mil.”

He looked down at the old lady’s face. She was still wearing the Coke-bottle glasses, the edges disappearing under her skin, the specs sewn in.

“Seems you were full of surprises,” he said, shaking his head.

Richter stepped around her and the shattered glass of a coffee table and headed past a thousand porcelain owls looking at him from the shelves.

In the back room where the drones had spotted the heat signature, the room was stuffed with clusters of Titan neurochips, 25 million cores a piece, knit together in tightly coupled supercomputing arrays in a liquid-cooled bath.

“Seems the old lady had a supercomputing cluster, tuned for ML inference,” said Álvarez.

He squinted at her; she wasn’t much of a tech head. “How’d you figure that?”

“You know me. I just shoot things. But Watson’s been geeking out on it all day.”

“Ah.”

“Watson, what do we got? What was she using it for?”

Watson’s face flared upon his virtual vision. He was talking, but Richter couldn’t hear him. “Can’t hear you.”

Watson mouthed something and then flickered away. Richter waited impatiently until Watson popped back up.

“Can you hear me now?”

“Yeah. What am I looking at?”

“Well, she wiped part of it before we got here, but she didn’t get far before we iced the deletion. Looks like she was crunching people profiles scraped from a lot of different sources. Social media. Underground lists. Dark web dumps from hacked DBs. She’s running a custom OCTO, multimodal, stacked Metaformer neural net—”

“A what?”

“Open source build of a multifunction LTM. But heavily modded with proprietary code and plugins, most of it pirated as far as I can see. The thing can learn anything. Good at lots of tasks. Visual. Reading and understanding. Relationship graphs. Video recognition.”

“Got it. So what’s it doing?”

“It’s tuned to track people. A few graph layers with enhanced relationship detection, lots of facial rec, personality profiling. It can pull up any life recorder and parse through the video and summarize it and—”

“So she can find anyone and get their whole life history, psyche profile, who they know, and where they go.”

“Yeah, that’s right.”

“It fits,” said Richter.

“Fits what?” said Watson.

“What I thought. She’s a Linker.”

“Thought those were a myth,” said Watson.

“Well, you’re looking at one, so obviously not. Good work. Get it all. I want to know who she was looking at most recently and over the last few months. Go back at least a year. Maybe two.”

“Yes, sir,” said Watson and disappeared.

“What’s a Linker?” said Álvarez.

“She finds people for people who want to do bad things. Like if you’re looking for new girls to kidnap out of the Czech Republic or Vietnam so you can turn them into hookers, she’ll find the ones who have no family and who got beat hard enough by their alcoholic fathers to give them low self-esteem.”

“¡Puta madre! The cartels have people like that—we called ‘em rastreadores, trackers, but they were a little more low tech and not—” She waved her hand. “—this fancy. They just go to poor parts of town and buy little girls from shitty fathers or grab the girls right there, and you never see them again until they show up at the bottom of a river after they got pounded on by dirtbags for two years in a back room.”

“This is more than that. That’s just brute force. This is surgical. The old lady could find you someone that’s almost guaranteed to be a nobody so no one comes looking for them if some John stabs her and they find her dead in a back room at some rat-infested hotel. But it’s more than that too. I had an old friend in the FBI, when it was still around. He tracked a Linker that found people to kidnap but whose families had a lot of dirty secrets so that they never called the cops. They just paid. He never caught the guy. Just found traces and rumors. A good Linker will know more about you than you know yourself. They can read people in an instant. Give them ten minutes, and they’ll be sure. They read all the little things everyone else misses. Pure sociopaths. To them, everyone’s a construct, a program that they can take over like a virus and control. They can pick out that big undercurrent of insecurity behind a thin veil of confidence or flaws in people’s beliefs that makes them prone to following orders and believing whatever counternarrative you want them to believe like it’s the only truth there is.”

“So what the hell was this one doing?”

“If I’m right, then we just found the key to the whole damn thing. I’m betting someone hired her to track down young, pissed-off kids who got nothing to live for and who are just violent and suicidal enough to fall into the waiting hands of a terror cell, looking to hurt a lot of people and needing some impressionable, disposable fanatics to do it.”

“Dios mío,” Álvarez said and made a quick sign of the cross.

“Now we just got to figure out who hired her. And we need to find them before they do whatever it is they’re going to do.”

You can grab the rest of it right here on Amazon.

Thanks for reading.

The Abundance Boom: A Memo from 2030

It’s 2030, and I’m writing this from a world that looks nothing like what the doomers predicted five years ago.

Remember 2026? When every third think piece was about the AI-driven jobless apocalypse? When serious people and fools alike were really convinced that Claude and ChatGPT and Seedance 2.0 were going to create permanent mass unemployment and usher in some kind of neo-feudal dystopia?

Yeah, that didn’t happen.

And if you actually understood how technology and economics work, if you’d actually paid attention to literally any previous technology transition in human history, you knew it wasn’t going to happen.

But here’s what did happen, and it’s way more interesting than the doom porn everyone was addicted to back then.

Jakarta, 2029

It’s March 2029, and a five-person team in Jakarta just dropped a film that’s competing with Marvel for box office dominance. Not competing in some indie-film-festival-participation-trophy kind of way. Actually competing. $340 million opening weekend worldwide. Ninety-five percent on Rotten Tomatoes. Three of the team members are under 25. Their total budget was $2.8 million.

The director used to work at a call center. The lead “cinematographer,” though that term barely means what it used to anymore, was a wedding photographer. They used video generation models that didn’t exist in any practical form in 2026, AI tools that let them create footage that would’ve required $200 million and Industrial Light & Magic five years earlier.

They acted in it themselves, using cardboard props to fill in laser swords and flying cars and then swapped it with advanced AI video editing software they cloned from more expensive software they couldn’t afford. The software got cranked out by one of the new micro software creation houses that have popped up all over India after the crash of some of the big Indian IT outsourcing houses. The coders didn’t go to the breadline. They went to work for the neo-outsourcing-insourcing houses that are booming. Back in 2026, people thought AI would replace cheap outsourcing but cheap outsourcing plus AI just boomed along with the insourcing boom back in the western world.

Here’s the thing nobody saw coming either:

Hollywood didn’t collapse.

The studio system didn’t die. What died was the price structure. What died was the stranglehold that a handful of executives had on what stories got told and how they got told.

By 2030, Marvel was still making movies, but their budgets had cratered from $300 million to $60 million because half the VFX work that used to require 500 artists in three countries was now handled by 50 people using AI tools that did the grunt work in hours instead of months. And you know what Disney did with those savings? They greenlit twelve other projects that never would’ve gotten funding under the old economics. Weirder stuff. Edgier stuff. Stuff that didn’t have to appeal to the broadest possible global audience because the financial risk was 80% lower.

The Jakarta team? They made their money and immediately spun up three more projects, hiring 35 people in the process. Because here’s what the doomers never understood: productivity tools don’t eliminate human work. They eliminate the barrier between human vision and execution.

We’ve Seen This Movie Before

Look, I get why people were scared in 2026. I really do. When you’re in the middle of a transition, everything feels unprecedented. Everything feels like this time is different. But this time wasn’t different. It never really is. It just looks that way when you’re knee deep in it and it all feels so new and terrifying and exciting.

I’ve been writing about and working in tech since the 1990s. I started working in an internet startup when everyone thought that was a dumb idea and the Internet was just a toy. “You’re wasting your life,” they told me. Really. I went to work in Linux when my recruiter told me all the jobs were in Solaris. I told him “Solaris won’t exist in ten years” and he looked at me like I had two heads. I saw panics over TV and video games and encryption and every other technology. People used to panic about Teddy bears and radios destroying the world. And yeah I’m serious. Look at those news clippings in the links. “Radio Blamed for Freak Weather.” “Radio Harms Children.” Radio induced insanity drove a preacher to kill his wife.

People love a good panic.

We eat it up.

And we always make the same goddamn mistakes: assuming that current limitations are permanent, that one thing changes and everything else stays the same, assuming we can’t and don’t adapt and just stay static, assuming that economic complexity can be reduced to “machines do X, therefore humans can’t do X anymore, therefore unemployment.”

It’s the lump of labor fallacy dressed up in neural network clothing, and it was bullshit then and it’s bullshit now.

The Deflation Nobody Wanted to Talk About

Here’s what actually happened to the economy, and why most economists missed it: AI triggered the most dramatic deflationary spiral in modern history. But it was good deflation, the kind driven by productivity explosions rather than demand collapse, and most people don’t understand the difference.

Between 2026 and 2030, the real cost of producing a professional-quality video game dropped by 85%. The cost of producing a feature film dropped by 73%. Legal document preparation: down 68%. Preliminary medical diagnostics: down 54%. Marketing content creation: down 91%. Software development: down 62%.

Now, if you’re an X addict stuck in 2026 thinking, you’re panicking right now. “It’s the end of white collar work!” Sure, it meant some jobs changed.

But here’s what happened to total employment in those sectors:

It went up.

Why?

Because when you drop the cost of making a video game by 85%, suddenly there’s room in the market for 50,000 new game studios instead of 5,000. When you drop the cost of making a film by 73%, you don’t get fewer films, you get an explosion of content, an explosion of niche markets, an explosion of previously impossible creative visions finding audiences.

The nurse making $78,000 in 2026 is making $82,000 in 2029, but her purchasing power is up 40% because everything got cheaper. She’s not just buying the same stuff for less, she’s buying stuff that didn’t exist or wasn’t accessible to someone at her income level five years ago.

That’s not a dystopia. That’s abundance.

What Actually Happened to the Jobs

Okay, but what about the displacement? What about the concept artists who spent 15 years mastering their craft, only to watch AI generate beautiful artwork in seconds?

Yeah, that was real. That was painful. I’m not going to blow sunshine up your ass and pretend it was painless. It wasn’t.

But here’s what actually happened, and it’s not what anyone predicted:

The coders and concept artists who just wanted to do one thing and who didn’t want to adapt or learn something new, the ones who tried to gate-keep, the ones who didn’t have any big picture understanding and just thought in silos, yeah they struggled. The ones who thought they could wave it all away with moral outrage found out that it wouldn’t go away any more than digital editing was going to go away because people who cut film with scissors screamed about it.

But the coders who understood the larger features and the UX and what they were working towards thrived. Generalists thrived. Cross-trained coders. Full stack coders. And the concept artists who understood storytelling, who could art-direct the AI, who could bridge the gap between “technically impressive” and “emotionally resonant” became more valuable than ever in the age of AI. By 2027, senior concept artists with strong vision were commanding higher rates than they ever had, because now they could execute ideas 10x faster with AI assistance, which meant they could work on 10x more projects.

The solo marketing consultant from 2026, the one who was grinding herself to death trying to do strategy, execution, analytics, and client management as one person? By 2027 she had a team of eight AI assistants who could pull ad stats, analyze them, write client briefs, concept ideas, do research and more. The AI tools handled the grunt work, data analysis, first-draft content, A/B test setup, which freed her up to do actual strategic thinking and client relationships. She spends more time meeting clients in real life, sharing a coffee and a funny story about her kids and theirs and how hard it is to juggle in the modern world. Her revenue went up 340%. Her stress levels went down.

The content writer? He originally got axed because some lazy execs thought “GPT can do it” but got hired back when his clients started realizing they couldn’t tell good copy from a hole in the wall. And all their ad click-throughs dropped to a .001% because the copy was generic garbage without someone to shepherd the AI. He now wields a friendly and helpful AI like a wizard with a Cursor for writing style platform that has editing, researching, fact-checking and proofreading agents baked right in and he’s closing more clients than ever and doing the work in a quarter of the time.

The game developer who used to spend 70% of his time on asset creation and 30% on actual game design? Now it’s reversed. He spends 10% of his time directing AI asset generation and 90% on making the game actually fun, which is the part that still requires human judgment and creativity.

And purely AI driven games that got generated on the fly? Turns out they mostly sucked and nobody wanted them. When a great story suddenly goes off the rails because the AI made a dumb decision, that’s not fun. Neither is getting stuck in a random corner of the board because the AI botched the physics. Turns out storytelling still needs a deft hand.

This pattern repeated everywhere. The routine stuff got automated. The human judgment, creativity, and relationship-building became more valuable than ever in the age of AI.

The Jobs That Didn’t Exist in 2026

By 2028, the fastest-growing job category was “AI integration specialist,” people who help businesses figure out which AI tools to use and how to implement them without screwing everything up. Median salary: $95,000. Required education: none of these people have computer science degrees. Most of them came from operations, project management, or customer service backgrounds.

Then there’s “synthetic media director,” people who art-direct AI-generated video, making sure it’s coherent, emotionally resonant, and actually serves the story. They’re making $120,000-$200,000. Their backgrounds? Film school dropouts, YouTubers, wedding videographers, former Etsy crafters who learned the tools and had good taste.

“AI personality auditor” is now a standard corporate role. “Training data curator” is a whole profession. “Prompt architecture consultant” is a thing, and some of them are billing $300/hour.

But here’s the wildest part: by 2030, an estimated 23 million people worldwide are making income, not full-time jobs necessarily, but real money, from what we’re calling “micro-studios.” One to five person teams using AI tools to create content, products, or services that would’ve required 50-500 people in 2026.

A two-person team in Lagos is making educational content that’s competing with Khan Academy. A solo creator in São Paulo is making meditation apps that rival Headspace. A three-person studio in San Diego is making indie games that are hitting top 10 on Steam.

The barrier between “I have an idea” and “I have a product” dropped to nearly zero. And when that happens, you don’t get fewer opportunities. You get an explosion of them.

Why Hollywood Didn’t Die But Changed Instead

Let’s go back to the film industry, because this is where the doomers were most confident in their predictions. “AI will destroy Hollywood!” “Actors and writers are finished!”

Here’s what actually happened:

In 2026-2027, there was a genuine rough patch. Studios started using AI for background characters, for crowd scenes, for establishing shots. Some VFX houses closed. Some junior positions disappeared. The doomers did their victory lap.

But by late 2028, something unexpected happened: the total number of people employed in film and TV production was higher than in 2026. Not lower. Higher.

Why? Because the cost of production dropped so dramatically that the market exploded. In 2026, about 150 films got theatrical releases worldwide. In 2030? Over 4,000. Not because there were more theaters, streaming had already won that war, but because the economics of production and distribution had changed so radically that niche films could find their audiences profitably.

A film that needed to make $100 million to break even in 2026 could break even at $15 million in 2030. That meant filmmakers could take risks. When you’re gambling with all the money it takes to make a film now you don’t take risks. You can’t afford to. The price of failure is corporate beheading.

So, if you’re an entertainment industry exec who wants to keep their job, you’re not going to gamble on the strange or the sublime. Risk is the enemy. Instead, you’re going to clone what already works. You greenlight “John Wick,” but with a chick. You order up “Fast and Furious 14: The Search for More Money.” You remake a classic, hoping nostalgia will cover for a lack of new ideas.

But when AI changed the way it all worked, it rewrote the rules. That meant weird, personal, experimental films could find funding. Which meant the monoculture started to fracture into a thousand microcultures, each getting the content they actually wanted rather than the lowest-common-denominator product that studios had to make when each film required $300 million in investment.

Yeah, the $300 million tentpole blockbuster became rarer. They flipped that money into a series because they could now produce Game of Thrones for the budget of a Miramax flick in the 1990s. The $20 million film became ubiquitous. And you know what? Most of those $20 million films were more interesting than the algorithmic superhero content we were getting in the 2020s.

The Scary Part (Because There Always Is One)

Look, I’m not going to pretend this was all sunshine and abundance. There were real casualties. The transition from 2026 to 2028 was messy as hell.

If you were a mid-level VFX artist doing rotoscoping in 2026, your job probably doesn’t exist anymore. If you were writing generic marketing copy or churning out stock photography, that work evaporated. If your entire value proposition was “I can execute this routine task reliably,” you had a rough few years.

Some people didn’t adapt. Some people couldn’t adapt, not because they were dumb, but because they had spent 30 years mastering a skill that suddenly wasn’t worth what it used to be, and learning a completely new career at that age is genuinely hard.

The safety nets helped. Most developed countries had implemented some form of transition assistance by 2029, though the U.S. was predictably late and half-assed about it. But even with support, the psychological toll of having your expertise devalued through no fault of your own was real.

I’m not going to minimize that. The micro story of creative destruction is often painful even when the macro story is positive.

But here’s what you have to know in 2026: the transition period is always the hardest part. Once you’re through it, once the new equilibrium establishes itself, things stabilize. And the new equilibrium is almost always better than the old one for most people.

Why I’m Writing This

I’m writing this in 2030 because I want you to understand something: the people who were screaming about AI-driven unemployment in 2026 weren’t stupid. They were looking at real disruption and real pain points. But they made the same mistake humans always make during technology transitions: they assumed that the immediate displacement was the permanent outcome.

They looked at short-term job losses and extrapolated to long-term jobless dystopia. They looked at specific tasks being automated and assumed entire professions would vanish. But tasks are not jobs. Jobs are the intricacy of a thousand little tasks and changing requirements and little intelligent decisions linking it all together. They fundamentally misunderstood how technology, markets, and human adaptability interact.

The intelligence boom between 2025 and 2030 followed the exact same pattern as every previous technology boom: short-term disruption, followed by adaptation, followed by a higher equilibrium with more opportunities and greater prosperity.

Yeah, the opportunities looked different. Yeah, some people struggled with the transition. But the net result, the thing you can see clearly from 2030 looking back, is that AI made us richer, not poorer.

The nurse can afford more. The teacher can afford more. The small business owner can afford more. Not because their nominal wages skyrocketed, but because everything got cheaper and their purchasing power expanded.

The aspiring filmmaker can actually make films. The game developer can actually ship games. The entrepreneur can actually start a business without needing $500,000 in capital.

That’s abundance. That’s what productivity gains actually look like when they ripple through an economy.

The Pattern You Should Recognize

If you’re reading this from 2030, you’re probably nodding along thinking “yeah, obviously.” But I want you to remember something: in 2026, a lot of people believed the opposite. They believed AI would create permanent mass unemployment. They believed we’d need universal basic income or face societal collapse. They believed we were heading into a jobless future.

They were wrong for the same reason the Luddites were wrong in 1811, the same reason the Population Bomb writer was wrong, the automation doomers were wrong in 1960, and the people predicting that ATMs would eliminate bank tellers were wrong in 1975.

They confused task automation with job elimination. They assumed human wants were finite. They underestimated human adaptability. They failed to imagine the new opportunities that would emerge.

And here’s the thing: five years from now, when the next technology breakthrough happens, quantum computing, fusion power, molecular manufacturing, whatever—people are going to make the exact same mistakes. They’re going to look at the immediate disruption and predict permanent catastrophe.

Don’t fall for it.

Technology doesn’t destroy prosperity. Technology creates prosperity by making things cheaper, by expanding what’s possible, by eliminating the barriers between human vision and execution.

Yeah, the transition is messy. Yeah, some people get hurt. But the macro trend has been the same for 200 years: technology makes us richer, creates more opportunities, and expands human potential.

The intelligence boom was no different.

And whatever comes next won’t be different either.

What This Means for You

If you’re reading this from 2026, and you’re scared about AI, here’s what you need to understand: the fear is normal. The disruption is real. But the outcome is going to be fine.

Not fine in a “everything stays exactly the same” way. Fine in a “the new equilibrium is better than the old one” way.

The skills that matter in 2030 are the same skills that have always mattered: creativity, judgment, taste, strategic thinking, emotional intelligence, and the ability to adapt. If you can do those things, you’re going to be fine. Better than fine.

The routine stuff, the “I can execute this task reliably,” that’s what AI handles now. Which means humans get to focus on the interesting problems, the creative challenges, the work that actually requires human judgment.

Is that a different kind of economy than we had in 2026? Yes. Is it worse? No way. It’s better. More opportunity. More creativity. More abundance.

The intelligence boom made us richer. And if you were betting against human adaptability and technological progress in 2026, well, I hope you didn’t bet too much.

Because the optimists were right. Again.

They always are.

It’s probably obvious if you read my blog, but I love writing.

And as LLMs got more deeply ingrained in my life and workflow, I’ve come to love writing with AI too. It’s completely changed how I work in my professional life and even my personal writing like this blog.

Now before you run screaming into the night, wondering if AI puked out 100% of this article, don’t worry. I hand-crafted 85% of it with a loving touch.

That’s because I still love the craft of writing. You’ll never catch me just telling the LLM to barf out a story for this blog from a prompt.

I read pure LLM text and as a thirty-year writing veteran I can instantly see why it sucks:

Too many block paragraphs. Too stiff. Too academic. Too generic. No flow. Points that make no sense. Too many em-dashes.

I read pure LLM writing and I instantly start ripping it apart and redoing 70-80% of it.

But it’s still incredibly useful if you adjust your expectations a little. Using AI well is really about adaptability. It’s turning off the noise and hype and seeing it for what it actually is, a brilliant but flawed technology that is beautiful and ugly at the same time. You have to work with it, warts and all.

And really it’s about you too.

When a new technology comes out, do you fight it or adapt to it?

Do you view it through a warped lens that lets politicians and media heads and X influencers define it for you, or do you dig in and experience it yourself and make up your own mind?

But that doesn’t mean AI tools are perfect or that they can or should do the whole job for you. If AI just wrote this whole blog for me, what’s the point? Why even write a blog? This is a place to share my thoughts and insights with the world, so I want an AI co-pilot, not a driver. I don’t need full self driving when it comes to AI assisted writing, I just need driver’s assist.

But it’s not easy getting the best out of AI. It’s easy to get lost in hype or just not understand what these little machines can and can’t do.

As long-time AI veteran Ben Goertzel wrote recently on X:

“1) If you don’t know what you’re doing in a certain domain [with AI], you can get interesting and valuable results, but they will be mixed up with garbage and you’ll have trouble identifying the garbage...

2) If you DO know what you’re doing in some domain, but are NOT willing to creatively adapt your work-style to the LLMs... then you won’t get much out of them except time-saving on some rote-ish tasks.

3) If you DO know what you’re doing in some domain and ARE willing to open-mindedly and creatively work with the LLMs to route around their shortcomings and leverage up their strengths, you can amplify your abilities and output DRAMATICALLY...though still with some tradeoffs.”

That’s really the heart of it. Are you a person who adapts or are you rigid and unchanging like an LLM itself, your mental weights frozen and fixed?

The beauty of being a human is continual and lifelong learning. Adaptability. Change. We can keep learning new things, keep adjusting, keep finding a new and better way to do the same things we’ve done in the past. That’s the real joy of life.

Even after thirty years, I still love sitting down and just letting the words flow.

But AI has completely changed how I research, edit, proofread and fact-check stories, as you can see in the videos below where I share bits of my process.

In that video I’m using new software, called Inkstone that’s coming out in January 2026. Think of it as Google Docs, a Perplexity style web searcher, Deep Research, a fact checker, and a proofreader and more all rolled into one. You can pop on the waitlist from the link.

I have early alpha access because I wrote a bunch of the software with a very small team. I wanted to have something that could edit my documents the way Cursor edits code. I got sick of copy and pasting between Cursor, Evernote, Google Docs, ChatGPT, Claude, a web browser and more and so I just put it all in one place.

Why?

Because writing is not just writing.

It’s note taking. Half baked ideas written down quickly. Scraps you captured on a walk last night. You have to move things around. These paragraphs are better at the bottom or in the middle or at the top. This section doesn’t hang together right. It needs some kind of transition sentences to bring it together.

There’s also research. You have to go to the web and find something. You do some research. Then do a little more. You save it in a note somewhere and pretty soon you have a giant stack of notes that you have to copy and paste into the article. None of it is organized.

I wanted a better way. And I wanted it all in one place:

Go fetch me some articles. Go do some research. Proofread this article. Organize it all. Make it searchable. Rewrite this section a bit. Move this section to the top. Put blank lines between every paragraph.

Once technology changes how you do things, you have to change the tools to match that workflow.

Technology changes the way we do everything.

Sometimes it makes an old way of doing something obsolete. We don’t hunt whales to dig the white gunk out of their heads to make candles anymore. We just flip a light switch.

But even when it makes things obsolete, it enables new possibilities. City life grew massively as electric light lit up the night and city-scale sanitation got better.

Other times tech makes something easier and more accessible. Before the printing press, every book had to be copied by hand, and that meant usually just religious books got copied. After the printing press, the variety of books and the amount of books exploded across the world, dramatically expanding human knowledge.

AI is just starting to affect the world in strange and wonderful new ways. Its biggest impact has been in coding, letting teams scale up and crank out code faster, even as they deal with new problems along the way.

When it comes to writing and AI, think about it like a hand-crafted table versus an IKEA table. Before IKEA, every table was handcrafted and that meant only a small number of people could afford them or they had to make the table themselves, which took a lot of time.

After IKEA everyone can have a good, solid table. Maybe not a craftsman level table, but a table that works and does what a table should do.

But IKEA didn’t make the craftsman tables disappear. We just got more tables.

That’s Jevon’s Paradox in action. Once something gets cheaper and more ubiquitous, we want more of it.

Maybe in the 1800s, you had a single well-crafted table. Now you have a kitchen table, a dining room table, side tables by the bed, an office desk, and more.

Too many people are talking about AI as replacing people and doing all the work.

But when it comes to writing, I don’t want AI doing all the work for me.

Just some of it. I want a co-pilot, a friend, an assistant, a ready-and-waiting editor on tap at any time. At least in my personal writing.

But business writing?

So much of what we write is repetitive and formulaic. Meet me here. That contract needs adjustment. Write yet another sales email.

AI can and should speed that up.

But when it comes to a novel or a blog post, you want hand crafted with just a dollop of AI help along the way.

You can have it both ways. Let AI do a little or a lot, depending on the circumstances.

And whether we like it or not, it’s best to lean into change so we don’t get swept away by it. If you still insist on cutting film with scissors when digital editing tools sweep the world of film, you’re not going to make it. You don’t want to be making candles when electric lights become reality.

But that doesn’t mean we have to do it the way everyone tells us to do it or that everyone does it the same way. You have to filter the process through your own self and make it your own.

Some people want a little AI. Some want a lot.

Your tools should adapt to you, not the other way around.

How AI Changes How I Work and Didn’t Change It Too

It took a bit to get comfortable with AI in my writing. I’ve been at this writing game for a very long time now: 35 years. I’ve got a workflow that works. It’s tried and true and frankly I didn’t see any need to change it.

At first I tried LLMs and thought, this is absolute garbage. It’s only good for people who can’t write at all.

But slowly but surely my mind changed. I thought, what if I go about this differently? What if I don’t think of it as doing the writing for me but doing all the stuff I don’t like about writing. Editing. Researching. Fact checking. Proofreading. I started to use it for all of those things. I used at as an idea bouncer-off-er, and more and more of my research happened while I did other things.

That’s when it started to click.

I started to weave AI in a little bit here, a little bit there. Look something up with Perplexity. Have Gemini proofread and do a draft update. Have Claude take my notes and write a few paragraphs.

Over time I developed an entirely new workflow. I figured out the strengths and weaknesses of these magical and frustrating little machines. For professional paid articles I’ve managed to cut my time from 5-8 days down to 1 or 2. That means I might make $350 an hour instead of $15 an hour. That’s because AI often helps me skip a few drafts and that saves a lot of time.

(Join Inkstone waitlist)

Of course, whatever the LLM first gives you won’t be exactly what you wanted to say. I tend to rip apart any draft the AI gives me and redo 80% of it. But sometimes it comes out with gems and bits I didn’t think of, and that’s amazing.

The key is, it’s something to work with. Something I can rearrange like Lego bricks. It’s a structure.

In writing, structure is the way the whole article works together from beginning to end. It’s the way you introduce ideas and strengthen your points as you go and drive it all to a powerful conclusion.

A writer can just glance at the structure of an essay and “feel” that it’s right or wrong. Something doesn’t flow here. This point is off topic. This whole section needs to get cut. This section needs to move up. There’s something missing here. That’s the wrong word choice. This sentence sucks.

That’s true for masters in many fields. Dutch psychologist Adriaan de Groot looked deep into the minds of chess masters in the 1940s and showed that great chess masters don’t actually think 30 or 40 moves ahead. They don’t even really think much beyond the next move. They’re looking at structure, at the board as a whole.

Author Joshua Foer wrote about it in his incredible book Moonwalking with Einstein: The Art and Science of Remembering Everything.

“It was as if the chess experts weren’t thinking so much as reacting. When De Groot listened to their verbal reports, he noticed that they described their thoughts in different language than less experienced chess players. They talked about configurations of pieces like ‘pawn structures’ and immediately noticed things that were out of sorts, like exposed rooks. They weren’t seeing the board as thirty-two pieces. They were seeing it as chunks of pieces, and systems of tension.”

I’ve always been slow with article structure. It’s the longest part of the writing process for me. I know how to start and where I want to go but can often get lost in the middle.

AI can set up the structure faster than I can. Much faster.

And that makes writing an article like working with a recipe book. It’s an idea machine. A great chef doesn’t need to follow the recipe. They just get ideas from it. They can just look at it and understand that this salt gets combined with this fat, acid and heat in this particular way and get too it.

AI is the same. It gives you ideas. It’s a sounding board. It puts together a structure and then you can rebuild it faster.

There’s an old joke in engineering. If you want the answer to a question, don’t ask an engineer, just post the wrong answer and watch a hundred engineers jump in to tell you you’re an idiot and exactly why you’re wrong. An LLM can do that for you in writing. It can give you a good wrong answer. And that saves a lot of time.

An LLM and tools like Perplexity and Deep Research can help you find and synthesize information fast. You’ve got to do a lot of research in an article like this. Web look ups. Deep reading. Quick skimming. Fact checking. Notes to first draft as you see in the video below.

(Join Inkstone waitlist)

On Writing Well and Writing with Well with AI

But AI is not a substitute for you.

I bristle up big time when I hear people saying AI will just be able to do anything we can do in a few years. It’s mostly nonsense to get big funding. There’s something missing with today’s AI and that’s fine by me. I still love to write and I don’t want it doing all the work for me.

Of course that’s my personal writing. When it comes to business writing I might let AI do 70% of the work and only write a bit of it. That’s fine too. Writing in business is a means to an end. It’s a protocol. It’s a way forward but not the end in itself. Get my team motivated. Get the schedule done. Talk to investors. An email to customers.

It’s the difference between that beautiful craftsman table and the IKEA table. Sometimes you do a lot of the work and sometimes little to none of it.

If you just need a table for the office, you buy the freaking IKEA table. You don’t have the time to sit down and lovingly build it from scratch like some 17th-century craftsman. There’s only so much time in life.

When it’s a business productivity thing, let the AI shoulder the heavy load. Let it do the research and plug it all into a big piece that you can edit. You’ve got to check its work. You have to click each link and make sure it’s a real link, but it’s still a lot faster, and that means more time to spend on your business instead of writing yet another doc.

What I have found over the last few months is that AI is strong as an editor and proofreader. It can take a messy first draft and get me further along.

I find AI is utterly useless with a blank page.

It’s obvious why: It can’t read my mind or figure out my unique style.

But it’s damn good with notes or a draft when it already has something to work with.

I’d encourage every young writer to avoid AI as much as possible while you’re learning though. Write the old fashioned way. Learn the craft. Put in the work. If you do that, you’ll be that much better off when you weave AI into your workflow as an editor, fact checker, brainstorming buddy, researcher, idea bouncer.

Or if you do use AI right from the start, take time to write the old fashioned manual way too sometimes. Force yourself to put it aside and learn the craft.

I love AI. It’s a fantastic tool and getting better every day. But there’s no substitute for learning to do something the hard way, no matter what it is in life.

Buying the best woodworking tools won’t make you a great woodworker. Doing woodworking every day will though and then you get upgraded tools that you know how to leverage.

And once you’ve got that, a better tool will make you that much stronger.

What AI Can’t Do - The Jagged Edge

No matter how much hype you hear though, LLMs are nothing like people.

There’s an old saying in robots that “things that are easy for computers/robots are hard for people, and things that are easy for people are hard for robots.” This is more true than ever now. Welcome to the jagged edge of LLMs, as Ilya Sutskever called it recently on the Dwarkish podcast. It’s when benchmarks don’t translate to the real world, and AI still makes mistakes. Like when it looked up the Dutch psychologist who wrote about chess and called him `Adrian` de Groot instead of `Adriaan`.

On some domains the LLMs are trained incredibly well. They know a domain inside and out, and they feel superhuman on it. I see this on backend coding tasks all the time. The frontier models can often translate my natural language into a solution fast.

With something like coding, the jagged edge is the LLM’s understanding of visual abstractions and frontend: CSS, React.

They’re all terrible at it.

As you watch an LLM struggle with things it’s never trained on, you go from feeling like “this machine is incredible, beyond anything a human can do and we’re all going to be out of a job” to “this is the dumbest fucking thing in the whole world, how could anyone think that this garbage will do anything we can do?”

The jagged edge cuts hard.

The biggest thing these systems can’t do yet is abstraction. What do I mean by that? If you get cut by a knife one time, you abstract the concept of “pain” and “sharpness.”

Now you don’t need to see 10,000 images of sharp spikes or pointy edge fences to know they’re dangerous. You know a jagged rock is dangerous instantly. One experience was all it took.

Today’s models give the illusion of being able to do this, but once you hit the jagged edge, the magic just evaporates instantly and they struggle.

Experience is the next level in AI, and it won’t come overnight.

Ilya Sutskever said we’re entering an “age of research again” and the next step is “learning from interaction.” AI needs to go beyond static data scaling into an era where systems interact with tools and environments, learn continuously, and are shaped by richer feedback than next-token prediction alone.

Like people.

But we’re not there yet. So for now you’ll just have to accept that AI is imperfect, and frankly that’s a good thing because it means you’re still in charge of where you want to use it and how.

The jagged edge isn’t a bug, it’s a fundamental feature of where these systems are today and once you get comfortable with it, it means you’ll wield these imperfect tools better and better and you won’t be swept away by hype and other nonsense.

You’ll just get things done.

Building Something Amazing Now

But here’s the thing: we don’t need to wait for perfection to build something valuable.

You understand the jagged edge now. And once you know that, you can use AI to level up. You aren’t affected by hype and nonsense. You don’t expect it to do it all for you. You don’t expect it to spit out perfection. You know its limits. Just like any tool you have to use it where it works best.

You wouldn’t use a hammer where you need a saw and a saw where you need a hammer.

And you can adapt. You can turn off the noise and ignore it. You can play with the machine and figure out what it’s good at, where it helps and where it’s useless. You understand its limits and you’ve learned to work with them, using it only where it makes sense.

That’s research. Web searching. Questions like “do human neurons do both memory and compute?” That’s fact checking. But you better double check it yourself too! That’s proofreading.

(Join Inkstone waitlist)

These systems are impressive autocomplete with superpowers in specific partial domains. Use them. Lean into their strengths. Work around their weaknesses. Don’t ask them to be general intelligence when they’re just table saws and don’t ask your table saw to be a hammer.

The future isn’t about AI replacing human judgment. It’s about humans who understand AI’s jagged edge wielding these imperfect tools with ever-increasing mastery.

That’s where the real power lives, right at the intersection of artificial capability and human wisdom.

And when you understand all that, you can surf the jagged edge, instead of getting sliced apart by it.

Something has changed.

Almost every senior engineer I know is writing very little code. Instead, models like Claude 4.5 Sonnet, GPT-5, GPT-5-Codex, and Cheetah are writing that code.

This is new. Very new. It wasn’t true six months ago.

It changed with the release of Claude Sonnet 4 in May. It went from, “this is kinda shit,” to “this is pretty good.” And then it really shifted with Sonnet 4.5 and GPT-5-Codex over the last month where it went to “this is pretty freaking amazing.”

These are not “vibe coders” shouting “this is wild” and “game changer” on every other X post about how you can write an “app” in ten minutes with a single prompt. These are serious engineers, people with 10 to 20 years of experience. And they’re shipping production code while writing pretty much none of it.

Back in the dark ages of six months ago AI couldn’t code its way out of a paper bag without you wanting to throw your laptop across the room in a rage. It’s only very recently that the world has shifted. But it has shifted.

Any engineer who insists that AI can’t code or only produces slop, or that they have to write 100% of their own code by hand, is going off an old mental pattern match that was true but is now dead wrong. That doesn’t mean that many amateur coders do not produce absolute garbage through their ignorance and pollute repos with trash PRs that don’t work. But experienced coders can now produce real, strong, working code with these little machines.

In March 2025, Dario Amodei, CEO of Anthropic said in “3 to 6 months, AI will be writing 90% of the code.” I pushed back on this. A lot of other folks did too.

Turns out he was right.

And he was totally wrong at the same time.

He was right because AI is writing 70%-100% of the code for me and many other engineers out there. But he was wrong because the implication was that it would be doing it without a human-in-the-loop.

In other words, we would just tell our little magic miracle machine what we want and then go do something else and then come back a few hours or days later and — bam — you got a working app! It will figure it all out on its own, make wise choices between challenging alternatives, set it up, test it and you can do more interesting things with your time.

That kind of automation remains a complete and total fantasy.

Despite classic Musk hype of Grok 5 having a 10% chance of being “AGI”, there is no model on the planet that is even close to this right now, for a thousands reasons. Take today’s model architecture and training techniques and algorithms and today’s data and hurl 200K GPUs at it, or 1M, or 10M, and the results will be the same: !== AGI.

The biggest reason is that even if the models get a lot smarter and more capable, they are not and cannot ever be, magic. The machine can’t read your mind or know what a good app actually is, unless it’s just a shitty clone of something that already exists. Only you can know that.

Without YOU, these coding bots are utterly and totally and completely worthless. Without you and me these models would just go off the rails and fail miserably. And yet, given a strong and well thought-out plan, that I have carefully designed and workshopped with the models themselves and with my team of real live people, these bots can and do work for a half hour and produce stellar code.

That is the paradox of today’s AI. The models are both freaking amazing and also completely idiotic at the same time.

I’ve watched Gpt-5-Codex code from a plan I laid out for two hours straight, correct lint errors, run tests, fix problems and then follow all of that great work up with a ridiculous mistake that a first day coder would not make.

My favorite was when it decided that the entire purpose of the Typescript deep research clone I was building (based on Langchain’s Python based Open Deep Research) was to research my test prompt of “find an under desk exercise cycle, in Europe. Pick the top five and cite evidence for or against.” After happily working for hours, it decided to hardcode that prompt into the search parameters for Tavily web search for every possible research question.

I screamed at it in all caps and rolled back the commit.

Look, two things can be true at the same time. Andrew Karpathy made waves recently on the Dwarkish podcast when he said today’s agents are trash and that real agents that can work like people are a decade away. I agree. True agents that can think, remember, plan, learn on the fly, adapt, socialize, gather context from external sources, remember shifting requirements and the big picture, create new ideas, and work like people are a decade away, if not longer or much longer. We need a different architecture and a different way to train these machines.

But again, so what? We still have little magic genies that are writing 70%-100% of the code! Today’s models are still spectacular and useful, even if they’re also maddening and idiotic at times too.

So I should be scared for my job then? I mean, it’s doing most of my job, right?

Except it’s not.

Because being a software engineer is 90% thinking and 10% writing code. So while from one angle it’s writing 100% of the code (actually more like 70%), from another angle it’s writing 0% of my code because without my skilled interventions and explanations it would have no idea what to build or why.

You also have to watch these machines like a hawk or they will fuck up and fuck up big time.

Sometimes I feel like a safety driver in a self-driving car, half asleep at the wheel, lulled into thinking this thing is super human and can do no wrong, and then it suddenly just veers off the road because it saw a shadow and I have to wake up out of a total stupor and grab the wheel back.

But at the same time, they’re good. Very, very good, and getting better all the time and so that’s why we’ve seen a shift where suddenly very senior engineers move into vibe engineering. They make extensive plans and then let the AI rip and test and refine while drinking coffee or making lunch or watching a podcast or learning French.

If this terrifies you, it shouldn’t.

You should be very, very excited, because you and your team are about to be able to do a lot more with a lot less. Think smaller teams shipping bigger projects, faster.

Many Paths to Victory

I will lay my cards on the table and say I love the new world. Absolutely love it.

Talking to a computer and having it do things has been a dream of mine ever since I started programming (badly) in C/C++ 25 years ago. Honestly, I’ve always hated programming. I’ve tried to get better at it a number of times over my life but it always felt too rigid, too inexpressive, too structured, versus the fluid fantastic and ever-adaptable world of human language.

Things have come full circle. Everyone is programming in English now and I’m here to meet the future where I always wanted to meet it. I’ve been ready for the “Hello, Computer“ era since I was a kid.

Just talk to the agent!

And you know what? My sense of coding has changed. I absolutely love it. I even love writing code by hand weirdly enough now.

So how does this all work? What I’m finding as I talk to more and more engineers is that everyone’s process is different. They find something that works for them and works well and they do great work.

My senior backend engineer breaks everything down into atomic tasks and makes sure only one thing is done in the space of a single context window loop. After that she compacts deliberately.

Ex-Vercel engineer and Typescript instructor Matt Pocock, does the same. He ruthlessly clears out context. In a reply, where I questioned why anyone wants a clean context he wrote back and said “The reason is all the research that shows LLMs perform better with fewer input tokens.”

Maybe the research does say that. But in practice, I rarely see the performance of the agent degrade because of ‘context poisoning.’ In fact, I see it get worse right after compacting when the model shows up like Mr Meeseeks with no freaking memory. I have to explain everything all over again. No thank you.

As for too much context? Sure. Sometimes, but not as often as people think.

I’ve developed a Spidey-Sense for when a Cursor or Claude Code thread has gotten too long and I flip to a new one but in general ‘context poisoning’ feels like straight nonsense, much like ‘model collapse’ last year. It’s something folks glom onto because it sounds good but in general it doesn’t play out in reality. It’s the kind of magic people believe because a few papers pointed it out.

And that’s the key for anyone coding furiously with AI today:

Don’t trust and verify.

There is no one path to victory. There are many.

Try lots of things. Be an explorer. Try every tool that comes out. But try it with critical thinking.

Don’t believe the hype, the research, someone on Twitter, or some article that says this is the only way to do it. These machines may not be AGI, but they are general-purpose-ish and that means they can do a lot of good work if you figure out how to work with them and their bizarre little quirks. Put your critical thinking cap on! Always be asking the questions:

• Does this match my experience?

• Does this really work this way?

• Why?

• Does it work at all?

• Is there a better way?

• How?

Don’t accept anything at face value. Prove it to yourself. Prove it in your own workflow. Prove it in the code.

My Workflow

So what is my workflow? It basically boils down to a few things:

1. Plan 80% of the time in pseudo-code/text.

2. Have the agent write code only 10% of then time and then test, test, test and bug fix the other 10%.

3. I also break out any commands I use over and over in my Flashbacker system for Claude Code.

Let’s break it down.

Working-shopping With AI

A lot of people think writing prompts is hard. As a writer for 30 years, I can’t relate in the least. Writing is my super power and describing what I want clearly and effectively is what I do.

I call the first phase workshopping with AI.

I tend to write out a long, rambling prompt of what I want to do, with links to anything that helps explain it. I then tell Claude to:

“Output your understanding of what I said. Output a plan to address it. Include a TLDR. Make no changes. Just plan.”

If I do have a magic prompt, that’s it.

I should probably have this as a slash command as I type it often enough. But I just type it as a force of habit.

It serves several key purposes:

First, it makes it immediately clear when the model does not understand what you wanted. If it doesn’t understand, you don’t need to read the rest of what it wrote. You can just prompt it again with a different/better explanation.

It’s very easy to get sucked into thinking these things are human and that they always understand you.

They do not.

And the faster you realize that and prepare for it, the better off you are.

Second, it includes a TLDR. Agents often write a tremendous amount of text very fast and reading it all is a great way to get exhausted in short order. So ask for a TLDR. Again, if it doesn’t make sense, you can skip reading the rest and re-prompt it.

Third, if all that makes sense you can then read the plan. You’ll spot other things. A file in the wrong place. Another misunderstanding. A duplicate function. Something it missed. Code that doesn’t need to exist because there is a better way to do it or a different way to do it, etc.

Once I have this I may prompt to refine it and tell it to “output the updated plan.”

I will then copy and paste that command into Cursor/GPT-5-Codex and say something like:

“This is what Claude thinks about what we are doing. Look at it with critical thinking and a critical eye. Tell me if you agree or disagree and cite evidence for or against in the code or documentation.”

I will then bounce back and forth a few times, doing my own docs reading, and code reading, and sharing that context with the agents, until I have it refined. I also tell Claude to strip out all the code that it hallucinated on the fly and just put it in with references in comments so that the actual implementation agent does not follow it blindly.

Basically, it all boils down to workshopping and refining a plan.

This is something you do in your mind’s internal dialogue or with notes or pen and paper without realizing it. Now you are externalizing it to give the agent proper understanding.

Capturing the Plan and Turning It into Tasks

After the plan is ready, I tell Claude to capture it into an issue using my /fb:create-issue command. It’s really just a prompt to tell it where to put that issue (docs/issues) and what should be in it. You could mod it to live in Git issues or whatever you like. I prefer it in a local file I can easily read. I then quick read it to see if it captures what I want and then I run the /fb:create-tasks command, another prompt that basically says decompose this into atomic, checkbox-able tasks. Those live in docs/tasks/{NAME_OF_ISSUE}/

Now that I have those tasks I will review them and if I see something is missing or needs to change. I will just change it if it is a small edit, or if it is a lot, I will tell Claude to update it with the /fb:/update-tasks command and then double check them.

They look like this:

Now that I have the tasks I can basically @ them in Cursor for GPT-5-Codex and say:

“Review all these and then output your understanding of what we are working on and output a plan to begin.”

I review it’s plan to make sure it matches up and then I just type:

“Do it.”

I can watch it code for hours this way.

Sometimes I use the new Cheetah stealth model in Cursor and it is blindingly fast. It feels like a diffusion model. It is ridiculously fast and can edit and read multiple documents at once. It is also stupid. So only give it very targeted, scoped tasks like import updates or doc updates. It can do those at the speed of light and will save you a lot of time. But let it write your new code and it is going to screw up.

But otherwise, pick a workhorse model like Codex and make that your main implementer agent.

Along the way I may need to stop it to tell it to do it a different way or because it came across something I didn’t expect, like a different library I need to do something or maybe I thought about it and I want it to return a different data structure or something else.

If that change starts to be major or makes us drift too far from the plan then I will stop and loop back to my planning phase and refine it then run /fb:update-tasks to capture it into the plan as a new phase.

Or I might just stop it and write it differently myself to get it done faster if I know exactly what I want.

Because the plan has checkboxes, the agents know to check them off after various steps, or I just tell Claude “look at the last commits and update progress.”

I run commit all the time with WIP messages. The more commits the better. Forget clean commit histories. The histories are there as a backup so you can rollback or unfuck what an agent did. For major commits, then you can push a better message that makes it clear this is for a human to read.

Commit too slowly and you are asking for trouble because agents can code faster than you. So they may successfully fix files that you didn’t check in and then screw something else up later and you are left to untangle it. Worst case you need to roll back the commit and now you just undid good work too.

So commit early and often. Like after every change.

Don’t Trust. And Verify

I also have a philosophy on testing that works very well. My basic rule if I cannot see it happening on a console log, or in a log file, or in a database, or a GUI interaction then it did not happen.

I cannot tell you how many times one of my agents wrote a test that did not show some background message and the output looked good but it was totally and completely FUBAR.

As I was writing my Deep Research agent I notice a message that was like “I can’t see what I am researching so I will ask the user what they want” because it was losing context on what it did because of a bad reducer that clipped the previous messages and retained only the last message. But the test did not show anything wrong because the Langgraph was finishing and the output showed successful tool calls. Meanwhile the agent was just reading the snippets from Tavily search and declaring itself done and failing to do any further research or web page reading on subsequent loops that always went to the end.

I repeat, if you did not see it and you cannot confirm it, it did not happen.

Prove it. Prove it to the AI and to yourself. Otherwise expect problems because the LLM will happily tell you something is “working” because of a bullshit/incorrect test it wrote.

Agents.md? SubAgents? Magical prompt Templates? Agents 2.0 and other Bullshit?

I’ve just started to do parallel agents a little bit but mostly I feel the same as Karpathy here:

“For example, I don’t want an Agent that goes off for 20 minutes and comes back with 1,000 lines of code. I certainly don’t feel ready to supervise a team of 10 of them. I’d like to go in chunks that I can keep in my head, where an LLM explains the code that it is writing. I’d like it to prove to me that what it did is correct, I want it to pull the API docs and show me that it used things correctly.”

And forget subagents. You can’t see what they are doing. Claude backgrounds them. They are good only for research but don’t let them make changes. That said, threads/tabs in Cursor or multiple Claude code agents with a different worktree? That works.

I use agents in parallel when I have a big, mature, tested codebase. But it is entirely bottlenecked by my ability/fatigue level to monitor them.

Monitoring 10? No way. That’s X influencer nonsense.

Trying to run multiple agents at the beginning when you’re working out what the code needs to do is a nightmare and will only cause you horrible, horrible problems. But once you have a standard and a clear refactor or a documentation update to do, agents are good at doing that kind of boring work in parallel without you paying super close attention.

They even make refactoring fun! And refactoring has never been fun.

Agents.md or Claude.md file? I have them. Couldn’t care less. Barely update them. People think these files are some kind of black magic. They’re not. They work fine (kinda, except when the agent just ignores them anyway) and if you like them, keep using them. But you’ll find that most things people hold sacred with AI is akin to superstition and folk beliefs. Mostly it’s not real.

How about MCP servers?

I use exactly three.

1. context7

2. sequential-thinking

3. chrome-dev-tools

All three are fantastic. Context7 and chrome-dev-tools are indispensable.

Context7 lets the model look up docs and dev tools exposes browser automation and screenshotting and Chrome’s developer streams to the the coding agent. If you are doing any work at all, you need context7 or the models will just make up types and methods that do not exist in a library.

If you do full stack or frontend work then Chrome Dev Tools makes Playwright look pathetic. It gives the agent the right context it needs to do strong work.

But I find most MCP servers to be utterly useless overkill. Frankly, I don’t like the protocol all that much and see it dying eventually. In 99% of the cases I don’t want my tool to run as a server. I can see why you would want it as a server. It decouples the code from the agent. You could have one MCP server in Go. Another in Rust. Another in Typescript.

Honestly, who gives a fuck? It’s annoying in practice and a nightmare in production.

My agents in production have a dozen tools or more, dynamically swapped in. I absolutely do not want to manage a dozen flaky-ass endpoints that could go down on me in production, especially if I’m vertically scaling them.

Much, much better to just write a simple tool with a cli interface and a good docstring to tell the agent how to use it. It can use `--help` and it’s all good. So make sure that cli help switch is, you know, helpful.

I actually hope Computer Use agents, that are undetectable, crush MCP servers and the pay-for-every-damn-service-via-newly-outrageous-API-pricing-that-used-to-be-free model dies a swift death. Just give me an agent that clicks around an interface with 99.9% accuracy. Or just give me a simple command line interface to an API or just the docs. That’s all you need. You probably don’t need an MCP server.

AI Slop, the Verification Problem, and AI Will Take All the Jobs (It won’t)

You’re probably wondering if the AI will take all the jobs in coding or if the code is any good?

The only people who think AI will take all the jobs are

1. terminally posting on X or

2. populist politicians or

3. Max Tegmark style doomers

4. sitting on the most clueless boardrooms in America and who think you just fire people and hire agents (all those stories about companies firing people and hiring “AI” are basically companies downsizing and saying “AI” to skirt other compliance rules. It’s not real in the way little green men on Mars are not real).

The reason AI can’t take any jobs now or anytime soon (if ever) is simple:

The verification problem.

What is the verification problem?

Simple.

If you’re already good at something, you can verify what the AI does. If you’re not, you can’t and you’re fucked and so is your company.

The hardest part of coding with AI (or doing anything at all with AI) is verification. If you’re using these systems for anything that matters, you need a verification pass that goes way beyond a lazy skim. That means detail-oriented human work — you must check every claim, every diagram, every link, every word, every line of code, every outcome and citation and fact when AI does something for you.

And who’s best positioned to verify?

The very people who are already good at whatever the AI is trying to do: the workers it’s supposed to replace.

Doctors can check medical claims. Senior programmers can check AI coding outputs. Strong copywriters can check that whatever GPT writes sings. They know a good turn of phrase when they read it and can make sure each paragraph flows from the one before it.

That’s the biggest irony of AI work. If you’re not already good at the task it’s doing, then you can’t tell if what it generates is good. You don’t have the knowledge or the context.

If you don’t know French, then you don’t know if a French translation sounds clunky or if you just told someone to eat shit in your new commercial because of new slang that sounds like the phrase you translated.

So I relate to the “learn to fucking code” posts on Hacker News. You should learn to code. Coding with AI made me want to learn coding and I did. So should you. Get better and better at it. Keep learning.

The irony is that despite writing very little of my code, I am a much better coder than I was a year ago. I’ve been studying coding like crazy, as much as three hours a day in certain stretches, just inhaling Youtube courses and O’Reilly books.

The more you know the easier it is to do something. Anyone who tells you “XYZ job is cooked” or “don’t bother learning this or that” or “don’t bother going to school” is a fucking idiot. Unfollow them immediately and never listen to another word that comes out of their mouth.

There is not a single thing in life that I regret learning. Everything I have ever learned has compounded and made me sharper and better at everything else I do.

A year ago I couldn’t write much (good) code other then some super complicated Bash scripts, which basically don’t count all that much. That doesn’t mean I never tried. Again, I started programming C/C++ in the 80s. And I’ve had various bursts of studying and coding along the way. I was just never very good. I was fantastic systems architect and sys admin but I just hated coding.

When I wrote The Joy and Pain of Coding with AI When you Suck at Coding back in April 2024, coding agents were awful in a thousand ways and it was like pulling teeth to get anything moderately passable written. The code I shared then is fucking embarrassing now. You can look at it and laugh. I do. It’s just utter dog shit that I would not ship today.

It’s also amazing that was a year ago! It feels like reading an article that I wrote twenty years ago. AI can help you learn in leaps and bounds.

That’s the journey of learning anything, of getting better over time. That’s Kung Fu. Hard work over time. And it’s amazing that my skills have leveled up so much in such a short period of time.

That’s what business necessity and pressure and three hours of intense code work a day will get you.

The harder you work and the more you have a real reason to learn it, the better you will get. AI coding is no different. Even if the machine is doing the grunt work, the more you know, the better you can make decisions. When you know what you’re doing, you know when the AI has just done something idiotic. If you can’t read or write code, you are up shit’s creak without a paddle.

But reject the “just write it all yourself” mentality. You don’t use a protractor when the calculator gets invented. Nobody uses a bow and arrow when they have an m4. the Mongols could shoot an arrow as fast as a bullet for short distances. You still want an m4.

If you insist on writing all your code by hand you will be in trouble in a few years when AI is doing work for most of your team. The new world is coming, ready or not. It will be as if you’re an analogue film editor, cutting film by hand and sneering at digital editing when computers came onto the scene. Eventually you’ll get left behind and be out of a job, not because AI took it, but because you decided scissors were all you need.

So keep learning. Keep adapting. Keep writing code, but get good with an AI workflow too.

The old world of writing every line of code is dying fast. We’re in the early adopter phase now and soon it will be early majority.

The new world is where sometimes you write the code, but mostly you don’t.

Maybe you think that sucks. It’s coming anyway, and any objections you have to it won’t stop it.

As Thomas Ptacek wrote in his awesome article “My AI Skeptic Friends are All Nuts“ there are no good objections that amount to rejecting the calculator for an abacus in your professional life.

But what about “you have no idea what the code is”:

To that he writes:

“Are you a vibe coding Youtuber? Can you not read code? If so: astute point. Otherwise: what the fuck is wrong with you? You’ve always been responsible for what you merge to main. You were five years go. And you are tomorrow, whether or not you use an LLM. If you build something with an LLM that people will depend on, read the code. In fact, you’ll probably do more than that. You’ll spend 5-10 minutes knocking it back into your own style. LLMs are showing signs of adapting to local idiom, but we’re not there yet.”

About half the posts on Hacker News to Peter Steinberger’s great article on coding with AI are “this is all AI Slop” or “he wrote 300K lines that could have been written in 20k.”

To that I say few things:

1. These folks are likely thinking about the models from half a year ago. They tried it a few times, for a short time, and then put it aside and went back to their old process.

2. They think of code as art or craftsmanship.

3. They’re worried about cost/their job.

To the AI Slop I point you again to developer Thomas Ptacek’s article “My AI Skeptic Friends are All Nuts“:

“But the code is shitty, like that of a junior developer,” he writes “Part of being a senior developer is making less-able coders productive, be they fleshly or algebraic...Also: let’s stop kidding ourselves about how good our human first cuts really are.”

And he goes further to the question of “but the craft” he says “Do you like fine Japanese woodworking? All hand tools and sashimono joinery? Me too. Do it on your own time.”

“I have a basic wood shop in my basement. I could get a lot of satisfaction from building a table. And, if that table is a workbench or a grill table, sure, I’ll build it. But if I need, like, a table? For people to sit at? In my office? I buy a fucking table.

“Professional software developers are in the business of solving practical problems for people with code. We are not, in our day jobs, artisans...Nobody cares if the logic board traces are pleasingly routed. If anything we build endures, it won’t be because the codebase was beautiful.”

Be responsible for your code. But refusing to code with AI will get you nowhere.

Lucky for us these machines can’t do taste or critical thinking or long term context or know what the hell you want to do in the first place.

Lucky for us, the real thinking of apps and life is asking questions and doing the hard work to figure things out. What do I want? Why? What should it do? How should it do it? What’s a good way to do it? What am I missing here?

People are still here. Again, all the scary stories you read in the mass media about AI taking all the jobs is just CEOs in an economic downturn of tariffs and trade wars and post COVID tech booms to bust. They are saying things to get them around worker regulations for letting people go.

Don’t believe the hype. There is not a single machine that can do anyone’s complete job today. These machines are fantastic task-doers and that’s it.

But long term memory? Long term context? Learning new tools and skills on the fly? Adapting? Updating priorities and project timelines? Forget it.

Most people don’t know what they want or how to articulate it. And what we want changes as we build something. You thought you wanted it this way, but discovered a better way as you worked on it. That is the process of creation. That is the process of taste and refinement and engineering and life! It’s iterative.

So jump in and embrace the calculator and leave your abacus at the door.

Find your workflow. Find your flow. And just go with it.

Keep coding. Keep learning. Turn the AI off sometimes and just code for the pure joy of it on a private project. Read a new O’Reilly book. Take a course. Always keep learning.

But use AI. Use it well.

A new life awaits you in the offworld colonies.

It takes a certain kind of personality to see a miracle and call it a monster.

We live in a time of miracles. If you're reading this, congratulations, you're a part of the luckiest 1% of 1% to ever live.

If you're alive today you're a part of a long chain of developments in human history where poverty has dramatically dropped, where people live longer and healthier and where hunger has fallen to single digits.

You can talk to anyone, anywhere on Earth for practically nothing via video or voice like you’re Dick Tracy. We have antibiotics that kill many of the bugs that usually killed us, MRI machines that can see inside of us and spot cancer before it takes over, and crumple zones in cars that make getting on the road safe. We have knives that don't rust because of the magic of stainless steel. We have refrigerators to keep our food cold and so it stays fresh longer and keeps us better nourished. Because of air conditioning we can live in places that were basically intolerable only fifty years ago.

But if you listen to the AI doomers it's all coming to an end and fast.

They believe that the entirety of human history and technological development, which has led to dramatically less starvation, less infant mortality, and eradicated many deadly diseases, will be reversed and go in exactly the opposite direction because of a dangerous new general purpose technology:

AI.

It doesn't matter that other general purpose technologies, other than weapons, such as the wheel, paper, the printing press, the internet, hammers and axes, have universally contributed to that dramatic uptick in lifespan/health/longevity/survivability. Those trends will not hold. Not only will all soon be inverted by this new general purpose technology, it’s even worse:

We're all gonna die! And soon! Wiped out like the Dinosaurs in a flash!

AI doomers are modern-day Chicken Littles, screaming about the sky falling. Their prophet Eliezer Yudkowsky just penned the modern equivalent of the Population Bomb, with the unbearably bleak title, "If We Build It, Everyone Dies," his fears built on a sand castle of pseudo-rationality and a steady diet of pulpy B movie sci-fi. He once called for bombing rogue datacenters by airstrike in Time magazine where he screamed that we should be less scared of "a shooting [war] between nations than of [an AI] moratorium being violated." In other words World War III is preferable to letting people build AI.

As reporter Nirit Weiss-Blatt writes in her new piece called the Rationality Trap:

“In 2024, Yudkowsky, told The Guardian that 'our current remaining timeline looks more like five years than 50 years. Could be two years, could be 10.” And “We have a shred of a chance that humanity survives.' MIRI president, Nate Soares, who co-authored the new book on how AI 'would kill us all,' told The Atlantic that he doesn’t set aside money for his 401(k). 'I just don’t expect the world to be around.’”

The book will probably sell well because fear sells. The media knows blood drives clicks. If it bleeds, it leads. Actually rational and sober analysis struggles desperately today to compete in an attention economy tuned for shock and rage clicks.

The doomers have been painting lurid pictures of gigadeath for years, with paperclip-maximizers turning us all into paperclips and malevolent superintelligences launching nukes. They’ve whispered their paranoid fantasies into the ears of politicians and journalists, demanding we slam the brakes on progress, regulate GPUs like they're assault rifles, and treat potentially one of the most transformative technology in human history as an existential threat.

(Clippy: Source Microsoft)

They warned that models trained on 10^23 FLOPs were too dangerous. Then 10^24. Then 10^25 and now 10^26. Today dozens of models sail past the 10^26 mark, from OpenAI, Anthropic and xAI, and yet civilization stubbornly refuses to collapse. In fact, it’s getting better, in ways both big and small, because of the very technology they want to strangle in its cradle.

The doomers are not just wrong; they are catastrophically, spectacularly, and dangerously wrong.

Their predictions have a 0% success rate and yet they keep getting press. They move the chains and promise us that this time they'll be right and we'll all be very sorry real soon, just you wait.

If their crazy policy suggestions get passed, like the MIRI report that calls for mass government surveillance of AI researchers and restricting GPUs like they're AK-47s, they will do real, lasting, nasty damage that's infinitely worse than the thing they're supposed to stop or prevent.

That's because when activists push big, scary headlines about the bad things they predict a technology will bring, a silent spring, mass unemployment, a new ice age, they usually dead wrong and they almost always ignore the good things we stand to lose without the technology: the jobs that never get created, the clean air we don’t breathe, the cascade of new inventions that never come to be.

On other words, technophobia has a body count.

When you throw a wrench in the wheels of progress, an alternative future full of opportunities vaporizes.

What diseases will we cure with stem cell breakthroughs decades later than we could have because we wasted eight years in the second Bush administration restricting the research? How many damaged lungs did we get because we killed off nuclear and kept right on burning coal to keep up with electricity demand?

Opportunity cost is invisible but its got one of the biggest body counts in history.

In science, every lost year compounds: a paper published in the journal of Technology and Innovation in 2011 estimated that the moratorium delayed first‑in‑human stem cell therapies for macular degeneration and Type 1 diabetes by at least five years. That's countless patients who could have regained their eyesight or thrown away their insulin pumps earlier.

But don't worry. While some people are out there selling panic and fear, the rest of us are busy building a better world. A world where AI isn’t a monster, but a miracle.

A world where, if we build it, society gets smarter and more resilient to shocks and better able to solve our biggest challenges and where AI sparks a medical revolution and more.

In short, a more intelligent world is coming and you shouldn't be afraid.

You should be very, very excited.

The Doctor's New Superpower

Take the story of David Fajgenbaum, a 25‑year‑old medical student. When he collapsed in 2010, he discovered he had Castleman disease, a guaranteed death sentence because there was no treatment. In an ICU in North Carolina, a priest read him last rites. The former Georgetown University quarterback endured four near‑fatal relapses in as many years and cycled through every available therapy.

Desperate, he turned research scientist. Fajgenbaum pored over his medical records, biopsied his own lymph‑node tissue, and ran laboratory assays. He hypothesized that sirolimus (rapamycin), an organ‑transplant drug never tested on Castleman's disease, could possibly shut down the disease. His clinicians agreed. The experimental treatment worked, and Fajgenbaum has been in remission since early 2014.

That's not an AI story yet. It's a human triumph story but it quickly turned into an AI story too.

Most rare‑disease patients lack an M.D. or a Ph.D. and a research lab, so in 2023 Fajgenbaum co‑founded Every Cure, a non‑profit determined to industrialize drug repurposing. Backed by a US$48 million ARPA‑H contract and a US$60 million Audacious Project grant, EveryCure is "developing the world’s most comprehensive open-source AI engine...for linking biological changes to human disease and therapies at scale...a 36 million cell (3,000 drug X 12,000 disease) heatmap with linkage scores for every possible drug-disease combination based on the strength of evidence. This will be open to all physicians, researchers, and patients to provide hope and spur new hypotheses."

TLDR, it finds FDA‑approved drugs that may treat overlooked illnesses.

(Source:EveryCure)

MATRIX has already:

• Flagged propranolol for angiosarcoma, an aggressive soft‑tissue cancer now in phase 2 trials.

• Suggested a three‑drug “Hail‑Mary” that placed a POEMS‑syndrome patient into durable remission.

Thousands of rare‑disease communities suddenly have plausible therapeutic starting points, doable because the drugs are shelf‑ready and safety‑profiled.

This is not speculative science fiction; it is happening today. And it exposes the fatal blind‑spot in the doomer narrative: when we wield AI to augment human insight, the default outcome is not catastrophe, it is lives saved.

And it’s not just about finding new cures in old data. AI is revolutionizing the very practice of medicine, giving doctors new superpowers to fight disease.

In a recent blockbuster moment for science and AI, the 2024 Nobel Prize in Chemistry went to DeepMind’s Demis Hassabis, John Jumper, and protein‑design pioneer David Baker for AlphaFold—the neural‑network juggernaut that cracked biology’s Holy Grail by predicting the 3‑D shapes of virtually every known protein, unleashing a torrent of potential breakthroughs from fighting malaria, to battling Parkinson's, to enzymes that munch plastic. Their open‑access model vaulted us decades ahead in medicine.

(Source: Google Alphafold)

In England, GPs are using an AI tool called C the Signs to scan patient records. This isn't about replacing doctors; it's about augmenting them, giving them a tool that can see the hidden patterns in a patient's history and test results. The result was a jump in cancer detection rates, from 58.7% to 66.0%, in a field where ever percentage point means more lives saved. That's more people getting diagnosed earlier, faster, and with a better chance of survival.

This is the reality of AI in healthcare: not a robot doctor making life-or-death decisions, but a smart assistant that helps human doctors be better at their jobs.

AI played a crucial role in the COVID vaccine and scaling mRNA research too. Now the same tech is in late‑stage trials for HIV, Zika, and a cancer.

Or how about the case of Google's Co-Scientist, a model designed to help scientists do their work better and faster. After professor José R Penadés and his Imperial College London team spent a decade working out why some superbugs are immune to antibiotics, he found that Co-Scientist came to the same conclusion in 48 hours.

"He told the BBC of his shock when he found what it had done, given his research was not published so could not have been found by the AI system in the public domain." Now the professor says it might have saved years of work.

"Prof Penadés said "I feel this will change science, definitely...I'm in front of something that is spectacular, and I'm very happy to be part of that...It's like you have the opportunity to be playing a big match - I feel like I'm finally playing a Champions League match with this thing."

From spotting the faint, almost invisible signs of future breast cancer in a mammogram years before a human radiologist could, to helping pathologists make faster, more accurate diagnoses, AI is weaving itself into the fabric of medicine. It's a scalpel, a microscope, and a diagnostic tool all in one, and it's powered by the same technology the doomers want us to fear.

But the revolution isn't confined to the hospital. It's happening everywhere. Let's look at where AI is making a difference in everything from engineering and software development, to the most basic of human needs: a place to call home.

Unlocking the Past, Building the Future

For decades, we’ve been told that building enough housing, especially in our most choked-off cities, is an unsolvable puzzle. A nightmare of zoning, costs, and endless delays. But what if we could design and build not just faster, but smarter?

In Oakland, California, a project called Project Phoenix is rising from the ashes of an old industrial site. It’s a partnership between modular construction pioneer Factory_OS and software giant Autodesk. Using a generative AI tool, they fed the system their constraints, studio, one and two-bedroom units, cost, carbon footprint, even quality of life metrics like sunlight and views. In hours, not weeks or months, the AI generated thousands of optimized designs for a 316-unit development, with a third of the homes designated for low-income or formerly homeless residents.

The real existential threats aren't rogue AIs; they're the slow, grinding challenges of energy, materials science, and engineering that limit our progress. For centuries, discovery has been a painful process of trial and error, a slow march of incremental gains. AI is speeding all of that up, getting us to solutions faster and faster.

Take the race for better batteries. We need batteries that are safer, charge faster, and use less of precious resources like lithium. The traditional approach? Mix and match elements in a lab, a process that could take decades to stumble upon a breakthrough.

Recently, Microsoft and the Pacific Northwest National Laboratory (PNNL) took a different approach. They used AI to find a new material for solid-state batteries. Instead of a handful of chemists, they unleashed a torrent of high-performance computing and AI models to analyze 32 million potential inorganic materials. The AI didn't just look for candidates; it filtered them for stability, reactivity, and their potential to conduct energy. In just 80 hours, it narrowed the field from 32 million to just 18 candidates. One of the most promising, may use 70% less lithium.

That's a big deal because lithium-ion batteries increasingly power everything from our grids, to our phones, to our cars, satellites and medical devices. Lithium demand is exploding and expected to rise 5x to 10x by 2030. Even worse, mining it is environmentally and geopolitically problematic so finding a formula that uses a lot less of it, while still allowing us to grow our civilization, to what augmented intelligence means for the world.

(Source: Our World in Data)

Again, despite the ludicrous headlines, it’s not about replacing human ingenuity but augmenting it, compressing 250 years of chemistry into a matter of months. This isn't a fluke; companies like Aionics are forming to build entire platforms to design and commercialize new materials at a speed that was unimaginable a few years ago.

Now, are these systems perfect? Of course not.

AI is a new kind of workflow. It shifts the bottlenecks to three new areas:

1. Problem definition/concepting

2. Iterations

3. Verification

If you can't define the problem, AI won't help you much. Iterations with an AI can be nearly instant or take a massive amount of time. The time it takes to get to a viable solution is often unpredictable and that can take speed ups from 10X to negative 10X in an instant. And verification still takes time. You have to run experiments in the real world of friction and dust. And if you don’t know what you’re looking at, because you’re already an expert, you won’t be able to verify the output. AI tends to make folks who already know something better at what they do because they can verify its outputs.

But these workflows will continue to get more streamlined and refined as we learn how to work with AI. And they will speed up our ability to design and build.

The goal isn't sterile, flawless perfection. The goal is to be better than the current speeds and time to value. To demand perfection before we deploy these tools is to condemn us to the slow, painful, and often fruitless march of trial and error.

We don't throw away a new drug because it has side effects; we work to mitigate them. Our first prototype build often looks like a Rube Goldberg machine and needs to be scraped but we learn from it and build a better mouse trap the second time around.

We embrace progress, flaws and all, and we iterate.

That is how every single human advancement has ever worked.

The Great Co-Creation

And what about the jobs? This is the doomers' favorite scare tactic, the one they trot out to frighten the masses. "The robots are coming for your jobs!" they shriek, painting a picture of a future where humans are obsolete, relegated to the scrap heap of history with chronic unemployment.

We've seen a flurry of stories like this. Derek Thompson just wrote a widely read story in The Atlantic that AI is killing jobs for college grads. We saw a flurry of copy-cat stories in the New York Times, Axios, PBS and the Guardian. Most of them, other than Thompson, didn't bother to question the AI is killing jobs narrative at all, they just ran with it, while screaming about the end-times.

But as blogger Noah Smith wrote in his Noahopinion blog:

"When people actually started taking a hard look at the data, they found that the story didn’t hold up. Martha Gimbel, executive director of The Budget Lab, pointed out what should have been obvious from Derek Thompson’s own graph — i.e., that most of the 'new-graduate gap' appeared before the invention of generative AI. In fact, since ChatGPT came out in 2022, the new-graduate gap has been lower than at its peak in 2021!"

These stories all come from fear and a failure of imagination, a fundamental misunderstanding of what this technology is. Humans are great at imagining all the jobs that might go away but they're terrible at imagining what kinds of jobs will get created.

We've already destroyed all the jobs multiple times in history. You didn't tan leather to make your clothes today or hunt the water buffalo for food.

But we've always replaced those jobs with new and more varied jobs. Nobody is calling for a return to whale hunting so we can dig the white gunk out of their heads to make candles. We flip a light and the electricity powers our energy efficient LEDs that last for decades and don't stink up the house.

We seem to resent progress today, to think that economics is a zero sum game where when we make things more efficient everyone gets poorer when it works in the exact opposite way.

It's a belief echoed by AI pioneer and socialist turned doomer, Geoffrey Hinton, who recently told the Financial Times "AI will make a few people much richer and most people poorer." He also said a decade ago that radiologists will be made obsolete by AI but we have widespread adoption of AI by radiologists instead and more radiologists than ever so I wouldn’t give his predictive powers much credibility at all.

(Source: Financial Times)

These kinds of beliefs stem from a fundamental misunderstanding of economics and progress in world history.

As Andrew Leigh writes in How Economics Explains the World, "In prehistoric times, the only source of artificial light was a wood fire. To produce as much light as a regular household lightbulb now gives off in an hour would have taken our prehistoric ancestors fifty-eight hours of foraging for timber...Today, less than one second of work will earn you enough money to run a modern household lightbulb for an hour. Measured in terms of artificial light, the earnings from work are 300,000 times higher today than they were in prehistoric times, and 30,000 times higher than they were in 1800. Where our ancient ancestors once toiled to brighten their nights, we rarely even think about the cost as we flick on a light."

And no, this time is not different.

If we have to bet on a black swan event that's never happened in all of human history versus countless evidence of adaption and integration of technology by people for millions of years, we should go with history. Of course, this time could might be different, because the past is never a perfect predictor of the future, but lets wait for actual evidence in the real world before we decide we need to take radical, extreme actions.

More than likely, AI isn't a replacement for human intelligence; it's an amplifier, a co-intelligence.

It's the beginning of a great co-creation, a partnership between human and machine that will unlock unprecedented levels of creativity and productivity.

(Source: Amazon and Ehtan Mollick)

Take chip design. For sixty years, it’s been a brutally complex puzzle. Engineers spend months painstakingly arranging billions of transistors, balancing power, performance, and area coverage. It’s a high-stakes game of Tetris at an atomic scale. In 2020, Google unleashed a reinforcement learning model called AlphaChip on the problem. Approaching it like a game, the AI learned to generate chip layouts that were superhuman, accomplishing in hours what took human teams months.

This isn’t science fiction. AlphaChip has been used in the last three generations of Google’s Tensor Processing Units (TPUs), the engines that power their AI services. NVIDIA is using similar tools, likely close to what they outlined in their Advanced Applied AI teams's CircuitVAE paper, to design circuits for their GPUs. This is how you scale a team and slash time to revenue. It’s not about firing engineers; it’s about giving them superpowers, letting them focus on architecture and new ideas while the AI handles the grueling optimization.

The repetitive, soul-crushing parts of our jobs? The endless data entry, the mind-numbing paperwork, the rote analysis? Let the machines have it. Let them free us to do what we do best: to think, to create, to connect, to dream.

We are not building a world without work. We are building a world where work is more human, more creative, and more meaningful than ever before.

People who embrace AI and work closely with it will displace people who refuse to work with it. Film editors who refused to switch to digital editing don't have a job anymore but there are more editing jobs than ever. That's because making films got easier with digital tools and that meant more films, which meant more jobs. Give people power tools and they are remarkably adaptable with them.

Who are the people who are best placed to get the most out of generative AI tools for film like Google's Veo 3 based Flow and other tools that control everything from character motion to camera angle?

You guessed it. Filmmakers.

They already understand shot composition and lighting and camera work. Artists are poised to get the best out of these tools like Runway Act 2 that ditches cumbersome motion capture suits for real time filters over actor performances. That means instead of actors and Hollywood being cooked, actors get a new lease on life and costs come down for making blockbusters.

We're already seeing a revolution in other areas too, especially software. When it comes to coding, we're about to have a lot more programs and software.

Agentic coding tools like Claude Code have taken software design to the next level. It's no longer just vibe coders using these tools. Real engineers and engineering managers are embracing them and moving beyond vibe coder 10x engineer hype to a squad of digital workers to augment their development process. Again though, it doesn't mean less jobs, just more software. As Naval wrote recently:

"If AI lets non-developers replace junior developers, imagine what it lets junior developers do."

AI levels up everyone and the more you know the more you get out of it.

Your engineering manager to can get things built now. So can the coder who spent two decades coding but needs to learn a new language or a new set of design patterns. Now they can adapt to new technologies faster with the help of AI and leverage their twenty years of wisdom along the way.

At the top of the chain, apps will get more powerful and more complicated as AI enters the workforce alongside traditional coders. At the bottom of the chain, small development teams will be able to do a lot more. And soloists will be able to create personalized programs and that will lead to some great one and five person companies.

We're entering an area of personalized coding. Dream up an app and build it for yourself, even if nobody else needs it. It doesn't matter if nobody else wants it because the barrier to entry will get so much lower.

People who can critically think, plan, strategize, reason and imagine will have a massive advantage in the coming world. Because now if know what you want and you know how to ask for it well, you can get a program I want out of AI.

Already, kids are coding with AI, like this eight year old. She'll grow up in a world where coding with AI is the default.

And if you tell her twenty years from now that people used to do it by hand and AI was once controversial, she will look at you like you have two heads and wonder what the hell you are talking about as she happily creates another app.

The Disconnect and the Reality

Right now, pessimism, especially about AI, is running rampant in western society and it's a terrible sign. When societies turn pessimistic and stop believing in the future it's often a self-fulfilling prophecy, as Johan Norberg shows in his fantastic book Peak Human. The book races through seven of history’s greatest golden ages—from democratic Athens to the tech‑turbocharged Anglosphere—to prove a simple law of civilization: open minds and open markets light the fuse for art, science and prosperity, while paranoia, protectionism and rigid hierarchies snuff the flame and send once‑mighty cultures tumbling into the footnotes.

(Source: Johan Norberg and Amazon)

Drawing on case studies where algebra sprung from Abbasid Baghdad and Renaissance Italy married art to industry, Norberg argues that progress and technological churn aren’t problems but the very oxygen of thriving societies—cut it off and stagnation, irrelevance and decline rush in.

Golden age civilizations embrace the future, embrace technology, embrace new ideas from anywhere and adapt them rapidly.

The doomers have focused us on phantom risks and hypothetical harms while ignoring the concrete, world-changing benefits. For every piece of AI-generated spam, there’s a medical breakthrough, a new material discovered, or a feat of engineering that was impossible just a few years ago. People who are obsessed with the imaginary monsters under the bed can’t see the miracle happening right in front of their eyes.

It doesn't mean there's no problems with AI. It's just not the problems doomers scream about. AI powered weapons and surveillance will supercharge authoritarian regimes and police states alike. But we won't stop nation states from building weapons. It's what they do, to protect their people and to fight other nations when things go wrong. AI can and will be used in weapons, but it doesn't mean AI is a weapon. Conflating the two is just a dishonest and idiotic rhetorical flourish to scare people. It's a lie.

AI is a general purpose technology. It's not a sentient being with its own desires. It’s a tool. It's a mirror that reflects our own ingenuity, our own compassion, our own desire to build a better world.

To fear it is to fear ourselves.

To demand we stop building is to demand we stop dreaming, stop striving, and stop reaching for a better future.

The future is not a predetermined path to destruction. It's a landscape of infinite possibility, and we are the architects. We can choose to be guided by fear, to huddle in the dark and curse the light, or we can choose to build. We can choose to use these incredible new tools to solve the oldest of human problems: disease, poverty, and ignorance.

So let the doomers have their bleak fantasies. Let them wallow in their apocalyptic predictions. We have work to do. We have a world to build. And in the world we’re building, AI isn’t the end of humanity. It’s the beginning of a new chapter, one where we are healthier, smarter, and more capable than ever before.

If we build it, everyone lives.

And we are building it, right now.

Imagine a very different 2025:

You can hop on a supersonic plane and get from New York to Los Angeles in 30 minutes or New York to Japan in 2 hours instead of 20. There is no climate crisis or activists gluing themselves to famous paintings. Stem cell therapies for damaged hearts came a decade earlier and saved tens of millions of people from early death.

We've got tremendous energy abundance with cheap, clean, nuclear energy and sweeping fields of solar cells with batteries. Machine learning solved protein folding 20 years earlier and powered a medical revolution, with permanent CRISPR cures for cystic fibrosis and sickle cell anemia. There are mRNA cures for breast and prostate cancer, nanotech robots that eat arterial plaque, robotic factories building everything from toys to advanced chips. A grid of 7th gen Starlink satellites beams down InfiniBand speed internet across the globe so a top doctor in New York can perform life-saving surgery via robot halfway across the world in ultra-high definition VR with no lag.

How did we get to this sci-fi future decades early? In our alt-history, the Luddites and the enemies of progress failed at every attempt to cripple, crush, or kill progress.

There was no backlash against "sonic booms" or runaway fears after the Concorde explosion. Instead, engineers did what they did with every other plane: they made them safer. As USA Facts tells us: "In 2022, there were 47 passenger injuries over 709 billion miles of air travel — you could circle the globe over 600,000 times for every one airplane injury."

Instead of hitting the kill switch on nuclear in the 1970s, the US kept right on building reactors and now we have an emissions chart like France, and so does the rest of the world. We've got safe, clean, abundant energy and there's no talk of degrowth or slowing down, just what to build next.

(Source: Our World in Data: French emissions chart over the last 120 years)

We didn't pull funding for stem cells in the Bush administration, and US and European activists never went on a pogrom against GMO in Africa after they already benefited massively from food abundance. Anti-nuclear activists didn't kill the American nuclear expansion in the 1970s.

Unfortunately, that's not the world we got. What went wrong?

The answer is that in way too many cases, the Luddites won.

(Ned Ludd the fictional leader of the Luddites)

They slammed the brakes on technology and progress out of blind animal fear and we all pay the price. The hidden price of technophobia is incredibly high. The real cost of these doomsday policies is in the air we breathe, the families who bury loved ones too soon, the new kinds of jobs that never get created, and the rockets that never blast off.

Activists always push big, scary headlines about what we're going to lose—a silent spring, massive unemployment, a new ice age—but they never talk about what we lose: the jobs that never get created, the clean air we don't breathe, the cascade of new inventions that never come to be.

When you throw a wrench in the wheels of progress, opportunities disappear in an alternative future that never happens. Enemies of innovation always think they're on the side of the light by slowing things down. But they never think about the side effects, about all the good things that we miss out on when we gum up the works. What diseases will we cure with stem cell breakthroughs decades later because we wasted eight years in the Bush administration restricting the research?

Opportunity cost is invisible—but it's one of the biggest bills in history.

The Clean‑Energy Future We Threw Away

Picture the United States in 2025 running on a French‑style nuclear grid: 70‑plus percent of electricity from zero‑carbon fission. The NASA Goddard/Hansen study calculates that the nuclear fleet we did build has already prevented 1.8 million premature deaths and 64 gigatonnes of CO₂ since 1971.

Now double or triple the size of our reactor fleet instead of activists and regulators strangling nuclear acceleration. We would have prevented another two to four million deaths and held temperature rises almost flat.

In short, there'd be no climate crisis and no talk of one.

Instead, we didn't break ground on a single new reactor between 1977 and 2013, and the first of them didn't go online until 2023, and two of the four were abandoned before completion.

Three Mile Island was the ultimate catalyst that scrapped the Atomic Energy Commission and replaced it with the practically anti-nuclear Nuclear Regulatory Commission. The NRC prioritized slowing down and throwing roadblocks up at every turn. Why? Because humans are great at overestimating big, flashy, short-term threats and completely failing to see long-term ones and what gets lost when we take one path versus another. It's easy to see a sunburn and realize you need more suntan lotion the next time, but getting lung cancer from smoking doesn't happen for decades, and so it's easy for people to puff away without awareness of what's coming down the pipe.

When it comes to Three Mile Island, nobody died. The worst dose at the fence line was less than a chest X‑ray. But that big, scary, near-miss event became the rallying cry that iced nuclear builds for forty‑six years. While the U.S. fought over paperwork, coal plants kept belching. Air pollution still kills every single year.

The body count of inaction dwarfs Chernobyl, Fukushima, and every other nuclear incident combined.

Ironically, you can blame the climate crisis on climate activists from the 1960s. Trace the results from their anti-nuclear activism and you get a direct connection to a warmer planet.

Again and again throughout history we've seen Luddites crush the future, and every time it means at its best lost time and opportunity, and at its worst, disease and death.

Eight Years in the Stem‑Cell Wilderness

Stem cells hold promise for treating everything from heart disease to badly burned skin. Heart disease kills 17.9 million people every year. Currently, there's no way to treat a damaged heart, to heal scar tissue, or regrow damaged parts of your heart. You're stuck waiting for a heart transplant. Transplants aren't scalable. There are only so many donors, and wait lists grow every year. Badly burned skin requires skin grafts. They also don't scale.

Imagine if we could regrow skin and heart tissue?

(Source: Swiss Medica)

Trials are already underway. In 2024, a University of Illinois cardiologist advanced stem cell therapy for heart failure to clinical trials, the first U.S. test of umbilical cord-derived intravenous delivery stem cells in patients for chronic heart failure.

But that's just the treatments of tomorrow. Stem cells are already used to treat everything from broken bones to blood disorders like sickle cell anemia and leukemia after chemo damages bone marrow.

It could have happened a lot faster.

In August 2001, President George W. Bush froze federal funding for embryonic stem‑cell lines. Overnight, America's most promising biologists became policy refugees. Labs shuttered or limped along on private grants; many simply moved to Singapore and the UK.

The embargo ended in 2009, but in science every lost year compounds: fewer post‑docs trained, fewer papers cited, fewer spin‑outs launched, fewer breakthroughs built on the back of other breakthroughs. A paper published in the Journal of Technology and Innovation on macular degeneration and Type 1 diabetes in 2011 estimated that the moratorium delayed first‑in‑human stem cell therapies for by at least five years, affecting countless patients who could have regained their eyesight or thrown away their insulin pumps earlier.

California tried to fill the void with a $3 billion bond, but even that couldn't replace a decade of coordinated federal effort.

Those delayed and missing therapies are an unmarked graveyard of vision, mobility, heart attacks, cancer, and time.

The Empty Fields of Africa

American and European activists enjoy organic kale while lecturing African farmers to avoid "Frankenfoods." The result? Vitamin‑A deficiency still blinds half a million children a year. Golden Rice—the GMO that fixes the deficiency—was ready in 1999. Twenty‑six years later, Philippine courts suspended rollout "out of caution," and most African regulators followed suit.

(Source: A scientist shows "Golden Rice" (R) and ordinary rice at the International Rice Research Institute in Los Banos, Laguna south of Manila, August 14, 2013. REUTERS/Erik De Castro)

Banana Xanthomonas Wilt (BXW)gene‑edited banana remains a massive problem for farmers across Africa, in lives and livelihoods. It slashes yields and sometimes kills as much as 100% of a farmer's crop. A single variety saves the crop, but adoption remains stuck below 2 percent because seed regulators haven't cleared it.

(Source: Wikimedia Commons, bananas ravaged by Wilt)

Smart bills to intelligently regulate GMOs get stuck in endless back-and-forth battles. Kenyan cassava, Nigerian cowpea, Ugandan banana—each sits in a regulatory holding pattern while real people skip meals. Genome‑editing could double smallholder yieldsget stucklate-2019 approval and erase entire pesticide classes, but smart bills to intelligently regulate GMO crops in endless back-and-forth battles.

A disease-resistant cassava in Kenya, a vitamin-A-fortified banana in Uganda, a blight-resistant potato in Nigeria — each sits in a regulatory holding pattern while real people skip meals. (Though in a recent glimmer of hope, Nigeria did make some progress with the of pod borer resistant cowpeas.)

According to a paper published in Frontiers in Bioengineering and Biotech, this is largely due to Africans' ingrained resistance to food innovations:

"Africa has historically struggled to adopt innovative agricultural technologies, which has significantly hindered efforts to ensure food security and improve livelihoods over the past century. A major obstacle in this regard has been the persistent skepticism surrounding the potential benefits of agricultural biotechnology.

The challenges contributing to this skepticism include a notable knowledge gap among stakeholders, widespread technophobia, or fear of technology, as well as inconsistencies with global agreements such as the Convention on Biological Diversity (CBD). Although these challenges are not exclusive to Africa, they disproportionately impact the continent, making the need for effective solutions even more urgent."

But it wasn't just homegrown fears. Instead of genuine safety concerns, what really stalled progress was an NGO-orchestrated scare campaign out of the US and EU. In his article A Dubious Success, the NGO Campaign Against GMOs, Robert Paarlberg, a professor of political science at Wellesley College, details how Greenpeace, Friends of the Earth International, and others recast safe, field-tested GMOs as "hazardous wastes" during the Cartagena Protocol talks — terrifying African delegates and trapping lifesaving crops in endless red tape.

It's the same story again and again in history. Activists trumpet the possibility of hypothetical risks while ignoring the certainty of hungry kids.

All of those are happening in the past, but what about the future of Luddite policies?

The mRNA Backlash

mRNA flipped the lights on during the COVID pandemic. Now the same tech is in late‑stage trials for HIV, Zika, and cancer. You'd think we'd be celebrating the potential end of these dreaded killers, but instead we have a small contingent of radicals fighting to .

How did it happen? Mostly because of Operation Warp Speed and the intense politicization around the COVID pandemic. Operation Warp Speed was one of the most incredibly successful technological acceleration projects in history, but it has the dubious distinction of being largely disavowed by both the left and the right — by the left because it came under Trump’s first administration, and by the right because of growing distrust toward health authorities during the pandemic.

But Operation Warp Speed was massively successful. It saw the US deliver a vaccine in 9 months and reach 40% coverage in the first year, instead of a decade.

As James Pethokoukis wrote in The Conservative Futurist:

"America was the first nation to roll out a vaccine. And it did so rapidly, getting the COVID-19 vaccine to the American people in less than ten months. And it wasn't just a single vaccine. Operation Warp Speed delivered three in record time...[D]rug companies had lost money during previous public health emergencies, such as the 2014 Ebola outbreaks, when treatments they developed turned out to be unnecessary. But now we needed those companies to make numerous big and bold—and risky—bets. So Operation Warp Speed placed orders for vaccines and therapies while still undergoing clinical trials, regardless of the outcomes. This encouraged pharmaceutical companies to expand manufacturing capacity so vaccines and therapies were ready to be distributed once they had the FDA's green light."

But it looks like instead of people cheering a promising HIV vaccine, we've got people fighting the most prominent trial because a few trial participants developed treatable hives. Who wouldn't take hives over dying of AIDS? Every known treatment has some kind of minor side effect. We push forward with them because the disease is worth fighting anyway. Instead, congressional firebrands drafted "mRNA moratorium" bills and NIH pulled co‑funding for the HIV vaccine program.

Freeze the pipeline today and you don't just delay vaccines; you kneecap the entire platform. Every postponed trial means the next generation of lipid nanoparticles, dosing algorithms, and delivery devices never arrive. That cascades into oncology, gene editing, even protein replacement.

The ladder of progress doesn't let you skip steps: kick out the first rung and you never reach the top.

We've Seen This Movie Before and the Great Chain of Creativity

Technophobia is the oldest genre in history:

• 1450s – Scribes smashed Gutenberg presses and chased printers out of town.

• 1840 – Painter Paul Delaroche saw the first daguerreotype and declared, "From today, painting is dead." It wasn't. It evolved.

• 1930s – The American Federation of Musicians tried to outlaw "talkies" because recorded soundtracks killed the pit‑orchestra gig.

• Robots have been about to take all the jobs for 100 years.

Every time, folks say "this time is different." Every time, it's a rerun. We adapt. We change right along with our technology.

We've always adapted our lives, culture, work, labor and business in the face of new technology. Technology isn't something that's outside of us. It's a part of us, a part of who we are as human beings. We're a long line of problem solvers and engineers. Ever since we fashioned sharp stones on sticks to make axes, we've constantly changed how we work and play and survive.

The tech rises and diffuses, culture adapts, creatives find new ways to be creative and humanity levels up.

Progress is recursive. Each step builds off the last step.

Once you have the ability to make clear glass, you get eyeglasses. Once you get eyeglasses, you get the microscope — a Dutch eyeglass maker created that modern marvel of science after seeing his kids playing with two lenses to magnify things.

A fabric maker then used the microscope to look at fabrics and saw a whole world of tiny creatures we didn’t know existed. Once you have an understanding of microbes, you get germ theory and the ability to fight infections.

Transistors enable personal computers; PCs enable the internet; the internet trains large‑scale AI; AI accelerates protein design; and protein design feeds back into better medicines and smarter biomaterials that improve the next generation of processors.

Break any link and the whole chain stalls.

"If not for funding for my early work in deep learning from the National Science Foundation (NSF) and Defense Advanced Research Projects Agency (DARPA), which disburse a good deal of U.S. research funding, I would not have discovered lessons about scaling that led me to pitch starting Google Brain to scale up deep learning. I am worried that cuts to funding for basic science will lead the U.S. — and also the world — to miss out on the next set of ideas," writes Andrew Ng in a recent blog post.

Kill basic research for a decade and you don't just lose ten years. Pass restrictive laws out of fantasy fears of AI killing us all and you do untold damage to the exponential curve.

When regulation works, it creates minor speed bumps. It doesn't set up concrete walls. It demands proof of harms instead of imagining a whole host of harms that haven't actually materialized in reality and may never materialize.

You make things safe by building them, testing them, breaking them, and iterating—never by outlawing the lab itself.

As James Pethokoukis wrote for the American Enterprise Institute (AEI), if U.S. regulation had merely frozen at 1949 levels, today's GDP would be $54 trillion instead of $27 trillion. That's two extra American economies left on the table—money that could fund universal clean energy, cancer cures, carbon capture, and Mars colonies.

Opportunity cost isn't hypothetical. It's the difference between ten million spinal cords repaired or not, between two extra degrees of warming or not, between poverty and plenty for half the planet.

That's why the next decade has to belong to the builders.

We didn't get flying cars because we chose paperwork over propulsion. It's time to choose differently.

The cost of Luddite policies is too high to let them keep robbing us all of the future.

The only thing we have to fear is fear itself.

We were promised flying cars and moon colonies. Instead we got idiots dancing on TikTok.

What went wrong? And how do we fix it?

But before we get there, let's just imagine a different timeline for a moment. What if things had gone very differently in the last 40 years?

First, think of all the incredible scientific and technological breakthroughs that came out of post-WWII America prior to 1970:

Nuclear power, satellites, the transistor, the Internet, television, water filters, cochlear implants, camera phones, cordless tools, information theory, moon rockets, scratch resistant lens, memory foam, freeze drying, solar cells, artificial limbs, and machine learning, to name just a few.

All of these technologies came out of the go-go techno optimism of the 1950s and 1960s. Many of them came out of the brilliant creative cauldron of Bell Labs, or from NASA trying to make sure astronauts and their equipment survived in unforgiving, cold, dark of space.

Now imagine many of them happening two or three decades earlier.

Even better, throw in all the technologies that almost came to pass that still haven't happened yet: supersonic travel, nuclear fusion, space travel, and nanotechnology.

They never came to pass because we never invested in them or cut off funding to them way too soon. But now imagine that we made those investments in the 1960s, 70s and 1980s. Instead of backing down, we went all out the way we did with the human genome project in the 1990s.

What would America look like now? What would the world look like?

Imagine an America powered by cheap, clean, nuclear energy and sweeping fields of solar cells with batteries, machine learning solving protein folding 20 years earlier and powering a medical revolution, permanent CRISPR cures for cystic fibrosis and sickle cell anemia, mRNA cures for breast and prostate cancer, travel from New York to Los Angeles in 30 minutes or from New York to Japan in two hours, nanotech robots that eat arterial plaque, robotic factories building everything from toys to advanced chips right here on our shores, and a grid of 6th or 7th gen Starlink satellites beaming down infiniband or even terabit per second speed internet across the globe so a top doctor in New York can perform life saving surgery via robot half way across the world in ultra-high definition VR with no lag.

What if we hadn't strangled housing development with zoning laws that made affordable housing functionally illegal so that it rapidly became impossible for the lawyer and the janitor to live near each in New York City or San Francisco? Instead of blue collar folks facing skyrocketing prices and forcing them to live an hour outside of the city and commute to work wasting time, money and energy, they'd live closer to work and walk there or hop on a quick train or bus.

What if we hadn't tangled up nuclear energy in endless environmental impact studies and approvals, and allowed anti-nuclear agitators and activists to raise objection after objection in courts, to that point that we didn't break ground on a single new reactor between 1977 and 2013 and the first of them didn't go online until 2023 and two of the four were abandoned before completion?

Instead of rising temperatures and pollution, we'd have a co2 emissions chart like France across the developed world, which didn't turn its back on nuclear and kept building after the 1970s and now gets over 70% of its electricity from clean nuclear energy.

(Source: Our World In Data)

As James Pethokukis puts it in The Conservative Futurist,

"To put it another way: if regulation had remained at its 1949 level, GDP today would be more like $75 trillion instead of a quarter of that. Imagine a U.S. economy nearly four times as big as the one today. Basically, we would have the U.S. economy of 2076 right now. Imagine median household pay of nearly $300,000 a year versus $70,000 currently."

(The Conservative Futurist)

In short, we'd have the economy of the tricentennial today and sci-fi technologies at every level of society. We'd have the makings of a Star Trek society, filled with technologies we can barely imagine.

If all that had happened would we even be talking about climate damage? Would we have populist movements and spin-dictator authoritarians on the rise all over the world? Would the idea of degrowth even exist?

Unfortunately, that's not the America or the future we got.

We didn't get unlimited clean energy and flying cars and Mars bases, we got people screaming at each other on Twitter and algorithms that promote the most polarizing posts. "The best minds of my generation are thinking about how to make people click ads," said Jeff Hammerbacher.

Instead of supersonic travel, fully automated factories, wireless infiniband and a nuclear renaissance, we've got a Black Mirror America, both in its mentality and the way its society is facing the future. Instead of more housing and robotic factories, we got degrowth, doomerism in every walk of life, fear of the future, the precautionary principal, populism, and protectionist tariffs.

And instead of people seeing science and technology as a powerful transformative force that brings a better tomorrow for everyone, they see life through a warped lens.

As Louis Anslow writes in his fantastic article "Black Mirror’s pessimism porn won’t lead us to a better future":

"Each new innovation gets an allegory: smartphones as tools for a new age caste system, robot dogs as overzealous human hunters, drones as a murderous swarm, artificial intelligence as new age necromancy, virtual reality and brain chips as seizure-inducing nightmares, to name a few. Episodes most often channel our collective anxieties about the future – or foment new ones through masterly writing, directing, casting and acting."

"Black Mirror is more pessimism porn than Plato’s parable, imparting to its audience a tacit lesson: fear the future more than the past. Fear too much technological change, not too little. It is an inherently populist narrative, one that appeals to nostalgia:

"Intellectually, we understand the present is better than the past in large part due to scientific and technological change, yet emotionally and instinctually we can’t help but feel this time in history is different, that the future can only get worse."

It seems no matter how much better science and technology make life we only see the future as getting worse in a strange kind of mass delusional hysteria that continually reinvents itself year after year. In years past everything from radios, to TV, to video games were rotting our brains and now its AI and social media. Tomorrow it will be AR glasses and whatever else we dream up.

Invent a better paperclip and someone, somewhere will be afraid of it and scream that it's the end of the world as we know it.

Progress has a massive ripple effect that benefits the whole world but we can't seem to see it anymore.

Don't believe life is better now than ever before in history? Just a quick look back at a few hundred years ago might change your mind.

Historian Michail Moatsos estimates that in 1820, just 200 years ago, almost 80% of the world lived in extreme poverty. That means people couldn't afford even the tiniest place to live or food that didn't leave them horribly malnourished. It means living on less than $1.90 a day in 2011 prices and $2.15 in 2017 prices. Actually you don't even need to go back that far. In the 1950s, half the world still lived in extreme poverty.

Today that number is 10%.

Nearly half of all children used to die, in every country on Earth, into the late 1800s, when economies were slower and more stagnant and less open.

In most first world countries infant mortality is now single digits or a fraction of 1% and even in the developing world it's 4%. On a recent trip to Japan I rows of tiny statues with red bibs on them. I asked someone about them and learned they were Jizō statues and the bibs often symbolized children who'd died.

To see just how stark that drop is, take a look at this chart which shows child mortality over two millennia. You’ll see that for most of human history it was a flat line of uninterrupted death and then it suddenly drops dramatically.

That's what progress does. It makes the whole world better. Each step builds on the last steps.

Once you have the ability to make clear glass you get eyeglasses. Once you get eyeglasses you get the microscope. A Dutch eyeglass maker saw his kids playing with two lens to magnify things and created that modern marvel of science. A fabric maker used it to look at fabrics and saw a whole world of tiny creatures we didn't know existed. Once you have an understanding of microbes, you get germ theory and the ability to fight them off.

Each layer of knowledge and tools enables the next level of knowledge and tools. When anyone can build on the past and abstract those lessons to new domains we get new breakthroughs and new ideas and new businesses. But when you stop making those tools or throw a wrench in that progress, the next level of technology stalls out too. You don't just impede the current technology, you break the great chain of creativity and stall out future technologies that would have been built on the last link in the chain.

When you read this article on your phone, you can thank the glass makers of Murano and the Corning folks who crafted super hard Gorilla Glass, along with the thousands of people who probed electricity through the years, from Benjamin Franklin, to Michael Faraday, to Nicola Tesla, to Edison, and all the invisible engineers and technicians whose names we don't know. You can thank the men of Bell Labs who gave us transistors and the Bell Labs offshoot company, Intel, who gave us the first powerful commercial microchips. You can thank the operating system pioneers and UI developers. And don't forget the thousands of factories all over Asia in a big, beautiful global supply chain, making smaller and more distinct parts, that feed larger factories that integrate those parts all down the line in over 100 countries. You can thank the massive container ships, the size of skyscrapers on their side, bringing you that phone from halfway across the world so it arrives at your door the day after you ordered it. You can thank the standardized metal box container with walls only three credit cards thick and floors that are antimicrobial, that protected the paper box with your phone as it travels from far away.

But we don't think about any of it. We pick up the phone, drive to the store, take the train to see our friends, eat in a bowl that doesn't crack, while looking up anything we want through our tiny magic window on the world. Tomorrow we'll just talk to the phone and have a brilliant built in assistant that knows exactly how to help us, like this demo of Project Astra from Google's recent conference.

Too often today, we find people who have a problem for every solution. Complaining is an international pastime. People scream on social media and gnash their teeth and they blame the media and rich people and the haves and the have-nots and corporations and politicians. But they don't offer solutions. They want to smash capitalism or kill all the lefties or crush all the conservatives but they don't have a better answer, just rage and delusion. To tackle the challenges of tomorrow, we need more thinkers and doers and inventors and less angry idiots.

The world is the world that we make, through our discoveries and creations and inquisitiveness.

But today Americans and most of the world have forgotten it.

We now fear the future and when you fear the future it becomes a self fulfilling prophecy.

Meanwhile, countries like China are racing into tomorrow with open arms. They've greenlit 10 nuclear reactors this year alone. American nuclear capacity is actually going down because we're retiring old reactors and not building new ones, while China's is on the rise.

(Source: U.S. Energy Information Administration)

Even worse, look at this chart about how people feel about AI on a country by country basis.

(Source: Ipsos)

Why are Americans afraid of the future when China embraces it?

The answer is simple.

Growth.

When you're country is growing, you're more optimistic about the future. When you're stagnating, you're fearful.

(Source: Pew Research Center)

China has experienced massive, breakneck GDP and quality of life growth over the last 50 years, an astonishing transformation that few people in the west can even imagine (more on that in a minute).

Meanwhile, America has seen roughly 2% growth since the 1970s, with a brief surge to 3% in the mid-90s into the early 2000s with the Internet boom. That's essentially stagnation when you factor in inflation.

People often look back on the America of the 1950s and 1960s with longing and it really was a very different era. We experienced 4% growth in the 50s and 60s and it was a time of astonishing changes from rapidly built highways, to home appliances everywhere, to astonishing growth in car ownership, to a dramatic rise in affordable home ownership across the land.

Most people alive today never experienced it and grew up in a time of slower moving change and so they've given up hope that life can be better. They lived in a time of higher prices, starter homes that cost $750,000, sudden layoffs and soaring grocery prices.

When you grow up in a time of stagnation, every disruption hurts and sets you back, whether that's COVID, inflation, the 2008 financial shock or massive layoffs. You're just trying to keep your head above water. It magnifies risks and makes them seem much bigger than your opportunities. The shocks hit harder because you have no cushion. You start to look at technology and progress as something that can take away what little you have left rather than something that can bring you a better life.

But if you lived in China over the last last 50 years you experienced something radically different.

In China today, their lives are dramatically better than their parents and their grandparents who lived through the horrors of Mao's China and the Cultural Revolution and the Great Leap Forward where hundreds of millions of people died. They've experienced a radical uptick in comfort, technology and security, the likes of which their parents couldn't even imagine in their wildest dreams. Soaring bridges, massive infrastructure scaling, new cities (they built over 500 new cities in the last 50 years many of which dwarf New York and London and Paris), new roads, 150,000 kms of train lines and 40,000 kms of bullet trains springing up from nothing while California spends 10s of billions and can't build a single bullet train for decades, direct electronic payments on their phones that skipped the step of plastic cards in between and so much more.

When you grow up in a fast moving, fast changing economy, if you lose your job, you just go work for the company that beat your company because business is scaling so rapidly, or you get a better job. The opportunities vastly outweigh the risks.

But in a slow moving economy, the risks seem to outweigh the rewards. It becomes a vicious cycle, where nobody dreams big anymore, nobody builds and nobody expects tomorrow to be better than today.

And that's what's happened to America.

We slowed down. We lost hope. We don't see a better tomorrow.

It's why today we don't have flying cars and clean nuclear reactors delivering emission free energy around the clock.

We have an America with a housing crunch where prices have gone up and up, where a young couple can't afford to buy an apartment or a car and start their working lives with 100s of thousands of dollars in debt if they went to college, where hard science innovations have largely stalled even as we saw a dramatic rise in computer and communication technologies during that brief spurt in the mid-90s to the early 2000s.

But even worse than all that, we have a growing chorus of doomsday prophets in everything from politics, to energy, to AI, with no solutions to our problems and who will only make things much, much worse if they succeed.

If we ever hope to turn it around, they must be stopped.

To build a better tomorrow we've got to start by smashing these insidious ideas before they poison more minds and spread like a nasty rot through the rest of society. Those ideas are a cluster of closely related ideologies, like degrowth, AI doomerism, anti-nuclear, and the precautionary principal.

Let's take a closer look at all of them and why they're bankrupt of any actual solutions but still seductively pull in so many with their potent cocktail of fear and apocalyptic visions like some kind of evangelical preacher sweating and screaming on stage as he looks to heavens and tells us fire and brimstone are coming if we don't repent.

Pessimist Porn and the American Decel Society

The reason they must be stopped is simple:

Broken ideas and delusional thinking never fix anything. They make everything worse.

It's like those old long division problems in school where you got the first step wrong and all the rest of the steps are wrong now too. If you can't even start off in the right direction you have no hope to get where you're going.

As growth slows down in any society, delusional ideas take root like poison weeds in the popular consciousness and given enough time they start convincing more and more people that terrible plans can actually work in the real world. If those delusional ideas get enough traction they can send a society hurtling through a tipping point that sees a country go into rapid decline.

Degrowthers tell us that we just have to "scale down destructive and unnecessary forms of production," to cut down energy use dramatically, and stop buying things we love. They imagine us all going back to the pre-industrial agrarian "utopia" of the 1800s and then we'll all be happy and free, working the land, dirt under our fingers, eating organic vegetables, the sun on our face. Never mind that, that was a time when half of all children died and that quality of life index was closer to Chad than Switzerland. Never let reality and facts get in a way of a good utopian vision.

AI doomers want to bring the Precautionary Principal from its home in Europe to America. The Precautionary Principal basically says "better safe than sorry." In other words, if you can't guarantee that there is no risk whatsoever you can't do something. If you can't guarantee that AI can never be used for anything bad then you can't make it. Of course, this is insane. It's a Catch 22. You can't prove anything is safe if you can't actually built it in the real world, or if research is so choked off that nobody is doing it at all.

We learn by doing. We get better by doing. We make things safer by making things in the real world.

We can't learn how to make a Boeing 787 Dreamliner safer by not building the Dreamliner. We have to build it and have a history of building earlier planes so we can learn from mistakes and failures. That's how things work in the real world and there's simply no way around it. You can't learn by not doing. You have to learn through experience.

Out of childish fantasy fears of AI rising up and killing us all these folks would rob us of DeepMind's billion protein fold database, the lightning fast development of the COVID mRNA vaccine, promising mRNA cures for skin cancer crafted by AI, already in trials, instantaneous machine translation when you're traveling to Japan or Italy or Peru and self-driving cars that are about 1000 times safer than human. 1.25 million people died in car wrecks every year. That's 3,287 deaths a day. Humans are terrible drivers. Even worse, more than 20-50 million more are injured or disabled.

But somehow AI is less safe?

It doesn't match reality in any way.

Beyond fears of AI itself, we have people screaming about how new datacenters for AI are "draining water supplies from the people who need them most," with scary pictures of dried river beds taken randomly from the internet that have nothing actually to do with those datacenters. We've got the same folks screaming about how datacenters are eating up too much energy in a criticism that borders on the farcical as datacenters account for a measly 1.5% of total energy use across the world and do useful things like make the Internet and their phones work so people can post their complaints about why datacenters are evil.

We have people like Ralph Nadar still fighting nuclear after fifty years and countries like Germany turning off their power plants so they can buy more dirty energy from Russia to fill the air with smog.

On the populist right we have politicians telling us with a straight face that 10% and 50% rises in taxes, aka tariffs, will save us and not just make the lives of everyday, working people and small businesses hideously more expensive. Maybe they think we can solve the housing crisis and soaring grocery costs by making everything else cost more too.

On the left we have people dreaming of a left wing "populist renaissance" so they can counter the radical plans of the right wing populists with their own equally delusional and idiotic plans.

Our science fiction is almost all negative across the board. We went from the Jetsons and Star Trek in the 1960s, glittering, hopeful futures, to the dark, rainy dystopias of Blade Runner and Terminator where nothing works and it’s all falling apart, or evil machines rise up to slaughter us all, or Black Mirror, where technology always has a hideous dark side and goes terribly wrong no matter what you do.

Instead of incredible scientists turned sci-fi authors and futurists like Isaac Asimov, we have futurists like David Wood, who instead of imaging a better future with powerful new technology that transforms our lives in wonderful new ways, is pushing extremist narratives like Elizer Yudkowski's "AI will kill us all" and excited that it will be the "most important book of the year" and not the Population Bomb of its time.

In other words, we're depressed, delusional and in decline, while the barbarians are battering at the gates.

This poisonous, hateful, hopeless mentality has taken over too much of America and much of the western world. We look inward and see nothing but problems and we look to tomorrow and see nothing but bad things coming over the horizon. It's swept away the can-do, big risk taking and bold mentality that made America the most powerful country on Earth. It's made people believe they won't have a better tomorrow no matter what happens, as if tomorrow is not built by our choices.

And it's happened at exactly the worst time, when China is on the rise. After the fall of the Soviet Union, we entered a world with America standing tall as the only true economic and military power on the planet. That world is gone. It's not going. It's gone. The Pax Americana is over. We already live in a multipolar world, with China the second biggest economy and soon to be the largest.

China now has the biggest car industry in the world. The largest fast food chain is no longer McDonald's, it's Mixue Bingcheng and it's Chinese, a company with zero restaurants in the US. 9 of the top 25 chains in the world are now Chinese. It has mega-cities, incredible infrastructure, multiple self driving car companies, the most dominant EVs in the world with 58% of the global supply, the biggest battery company on the planet by a huge margin in CATL powering everything from cars to national electric grids to drones, the most advanced manufacturing in the world, drone food delivery and other sci-fi marvels. If your idea of China is a bunch of people on bicycles in Beijing, that world hasn't existed for decades. Now Chinese people drive motorbikes or take autonomous trains. They have more than 50 nuclear reactors planned. They're harnessing wind and solar with batteries.

(Source: BoFA Global Research)

They've gotten there with bold thinking and big risk taking. They've gotten there by doing what America once did best and did alone:

Building.

Dreaming big and building.

In other words, it's exactly the wrong time in our history for the American deceleration or "decel" mentality to take hold. It's a time when we need to lead on the world stage, not receed. It's a time when we need to innovate faster and build bold new technologies or we risk being overtaken by China and at worst ceding the future to them.

We won't stop China by trying to choke off their supply of AI chips. We won't beat them with tariffs and increasingly arduous chip controls and by blocking their cars from America. The rest of the world will happily buy those cars.

We can only win on the world stage the way we always have, by building a better mouse trap. Make things that people want and need. Do it better than everyone else.

In other words, compete.

To do that we've got to fix our mindset and know who the real enemies are:

Ourselves.

And to do that, we just have to remember who really makes the world.

Tomorrowland, Decay and Why Money Does Equal Happiness

So who makes the world?

If you trust the history books, it's generals and world leaders.

In school, our history books are filled with pages and pages of war mongers and villains and killers. We read about civil wars, and Stalin and Hitler and Mao, people who killed hundreds of millions of people and drove their societies to disasters.

These are the people we should forget, not the people we should remember. Destroyers. Murderers. Misery bringers.

The don't make the world, they fuck it up.

It's builders and inventors and entrepreneurs who really make the world. We stand on the shoulders of those giants.

The world we live in today is the result of countless breakthroughs and ideas made real by the people of the past. We tend to think of history as constant wars, punctuated by short stints of peace but it's actually the reverse, long periods of peace, punctuated by short periods of violence.

And in between all those wars and breakdowns, the people of the past were making the world you live in today.

Take something a simple as a teapot. You probably don't notice it. I didn't until one fateful Saturday breakfast where I suddenly stopped and stared at it.

It was as if I was seeing it for the very first time. Its smoothly curved and gleaming surface mesmerized me for a moment.

It glittered in the early morning light as my wife I drank our Earl Grey and talked about the world and about all the wonderful things we wanted to do on our day off. All the sound drained from the room and it was just me and the teapot for a few seconds that felt like minutes. We'd bought it while living in Berlin and carried it with us around the world. Inside, at the bottom, it's discolored, having taken on layer upon layer of thousands of pots of tea, but outside it's still flawless, true to its name, stainless steel.

We don't think about the thousands of years of trial and error that went into making stainless steel. It's just a part of our life, serving us faithfully, asking for nothing, never calling attention to itself. It's clean and shiny and feels nearly indestructible. The secret ingredient is chromium. It creates a special protective shield of chromium oxide when it mixes with carbon and iron atoms of steel. That give it that gleaming shine and its invisible barrier against stains and rust. The chromium oxide keeps the ravages of moisture away from the powerful inner core of metal and carbon, deflecting stains like a kung-fu master deflects punches and kicks.

That stain blocking power is also why stainless steel replaced ceramic in sinks. Ceramic soaks up stains like paper soaks up ink, but that curved and sleek steel sink deflects everything from food, to bleach, to strawberry juice, to harsh chemicals, to ink, and more, never losing its shimmer.

Maybe you've noticed that your fork or spoon doesn't actually taste like anything? Go ahead. Put it in your mouth and see what it tastes like. Nothing, right? That's also because that magical chromium barrier. It keeps the metal from interacting with your saliva which corrodes steel by itself. As materials scientist Mark Miodownik writes in his amazing book Stuff Matters, it means "we're one of the first generations who haven't had to taste our cutlery."

Metals of the past soaked up all the little flavors from meals gone by, subtly distorting the flavors of everything over time, no matter how hard people scrubbed them.

But it's not just the material that matters.

It's the ingenuity and the genius that went into everything we see around us.

We live in a world where people before us solved countless problems so that we don't have to solve them. Every teapot, every window is a solved problem. Every smart phone, every car, every wooden floor, every steel and cement skyscraper, every refrigerator, every toaster, all those keys and locks, those closets you keep things in, the internet that wings these words to you around the world, the air conditioner keeping you cool, all those and a million more little things are solved problems. The people before us live on, infused into the very soul of the objects we touch and taste and walk on every second of our lives.

Layer, upon layer, upon layer of pre-baked solutions surround us every day. We don't notice them at all. They're just a part of the tapestry of our world. Every time we call someone on our phone, or get on a bus or a train, or drive somewhere, or put on clothes and shoes, or shove a delicious morsel of food into our mouth with a stainless steel fork, we're living off the genius and ingenuity of our ancestors.

And it's the builders that got us here. The scientists. The thinkers. The people who risked everything to start a company and grow it and bring that magical product to you. It's the builders that make life marvelous and that made America a super power and every other super power before it. The risk takers. The change makers.

We didn't get to be the most powerful country on Earth by strangling every new technology in red tape or by trying to make sure that nothing ever goes bad. We did it by growing. By taking risks. By exploring and experimenting and creating new and wondrous things that transformed the world around us.

A healthy society is a growing society. It's a society that creates new things in a constant stream. It's a society whose cities are getting larger and where more and more of that society's people have their needs and wants satisfied.

A society that's stagnating or shrinking is a sick society.

It seems simple and obvious but it's not obvious to many people. Builders have always had enemies and resistance but today, they're legion, a growing and powerful chorus of delusional people who fight progress and growth every step of the way.

Here's the opening paragraph of Less Is More: How Degrowth Will Save the World, by Jason Hickel, a degrowther bible:

"Perhaps it seemed like a good idea at the time: transfer land to big companies, rip up any hedges and trees and plant it all with a single crop, spray it from aeroplanes and harvest with giant combines. Beginning in the middle of the twentieth century, whole landscapes were remade according to the totalitarian logic of industrial profit, most of it for livestock feed, with the goal of maximising extraction. They called it the Green Revolution but, from the perspective of ecology, there was nothing ‘green’ about it. By reducing complex ecological systems to a single dimension, everything else became invisible. Nobody noticed what was happening to the insects and the birds. Or even to the soil itself."

What he is talking about is the Green Revolution and what actually happened is it saved billions of lives and earned its creators the Noble prize. What was "lost" was mass starvation and famine.

That was back when the 1968 book, The Population Bomb, predicted that we'd have no choice but to let 2 billion people starve because there just weren't enough resources to go around. Mostly we'd just let everyone in Africa and the Middle East and most of China die and there was absolutely nothing we could do about it.

Instead we got the Green Revolution and malnourishment is at all time lows across the world.

(Source: Gapminder)

The green revolution was a revolution in farming techniques, from cross breeding resilient plants, to pest control, to soil sustainability.

It marked the Third Agricultural Revolution, where we saw massively increased crop yields and agricultural production across the world. In the late 1960s, as Professor Ehrlich was penning the Population bomb, farmers and scientists were getting to work, using fertilizers, high yield wheat and rice, pesticides and irrigation.

Norman Borlaug, the "Father of the Green Revolution”, received the Noble Peace Prize in 1970. Yuan Longping did the same for rice, saving as many or more people in Asia.

And yet degrowthers describe one of the most import scientific revolutions in history as a failure.

While there is something in Hinkle's appeal for more diversified crops, and planting a garden in your backyard, the truth is that's for rich people in California who can afford to eat vegetables from small, local farms that are five times as expensive that don't have high enough yields to feed a hungry planet. The rest of the people on the planet have to eat too. Meanwhile saying "lets have less people" is basically on par with supervillain Thanos' "they call me a madman" speech in the Avengers who wanted to snap his fingers to wipe out half of life in the Universe as a kind of aggressive population control.

Super pessimists like Ehrlich are still an unrepentant pessimists to this day. In a 2020 interview, he said “There are too many super-consumers on the planet." It seems cognitive dissonance has little effect on the man, even at 88 years old as he was during the time of the interview. Organizations like the World Clock still push the population bomb idea but now conflate it with fears of rabid capitalist overconsumption exhausting all the resources on the planet.

Doomerism never dies, it just changes its talking points.

Even worse, the doomers are wrong again. A second Green Revolution is on the horizon in China right now.

As Pethokoukis writes in the Conservative Futurist "Chinese scientists report that by giving 'a Chinese rice variety a second copy of one of its own genes, researchers have boosted its yield by up to 40 percent,' according to Science [magazine]. Researchers at the Innovative Genomics Institute estimate that CRISPR gene-editing technology could improve plant photosynthesis between 25 percent and 50 percent. No doubt the starvation countdown clocks will continue counting down no matter what other green shoots of agricultural progress emerge."

Or as authors Ezra Klein and Derek Thompson write in their fantastic book Abundance:

"To the extent that degrowth has a specific climate plan, it is to shut off or scale down areas of production it deems destructive, like military investment, meat and dairy production, advertising, and fast fashion..."

"We know what it looks like when governments face the political fury of rising energy prices or fuel rationing. In 2022, ninety countries and territories experienced often violent protests over the rising price of fuel between January and September, according to a BBC analysis. In Sri Lanka—a country that Hickel holds out as a model for degrowth development—those protests led to the collapse of the ruling government."

When you slow down, you don't get shiny, happy people, working the land with smiles on their faces, you get food and fuel riots or famine.

(Abundance)

Despite all the evidence that happy societies are rich, growing societies, that hasn't stopped a whole host of people throughout history from seeing growth as a disease that needs to be stopped by any means necessary.

We've already seen a backlash to the Abundance movement on the left as some folks dream of a "populist renaissance" to match the populism push on the right, as blogger Noah Smith wrote recently:

"In a recent discussion, Klein pressed two of his progressive critics — Zephyr Teachout and Saikat Chakrabarti — to tell him what concrete problems they think an antimonopoly movement would solve. Again and again, Teachout’s answer is “power”:

It’s a democracy vision…I think for 40 years…we basically stopped asking the power question…[B]oth Republicans and Democrats got on board with…this idea that we should just focus on outputs and not on power. So that’s part of the reason you hear some resistance from the antimonopolists to your [abundance] vision.

"Here we see the fundamentally different goals of abundance liberalism and anticorporate progressivism. Abundance liberals care about what stuff people get, while anticorporate progressives care about who holds power in society. The progressives have trouble explaining exactly how changing the distribution of power would lead to better material outcomes for the masses, but that doesn’t phase them much; to them, reducing corporate power is an end in and of itself."

Noah goes on:

"As I’ve noted before, this ideas smacks of class resentment — a professional class that cares more about dunking on the entrepreneurial class than about helping the working class. That’s why critics of Abundance, like Aaron Regunburg, tend to focus so strongly on accusations that abundance liberalism is being secretly supported by class enemies:

Our concern is that corporate-aligned interests are using abundance to head off the Democratic Party’s long-delayed and desperately needed return to economic populism…[W]ho are the villains we should be naming? A growing number of Democrats are coalescing around a simple answer to that question: oligarchy…[M]any Democratic elites still oppose any attempts to identify billionaires and corporations as villains…[T]hey are terrified of the prospect of a populist takeover of a party…that has for decades served as a comfortable partner to oligarchy…"

They're right to be terrified of that takeover because it's an ideology bankrupt of any actual solutions. Essentially the response of degrowthers to "let's have a richer, more prosperous, more stable, more green society by building and cutting back overzealous legislation" is "we don't like billionaires and they're evil," a kind of American Black Mirror of Maoist ideas.

Let's be clear, there is no "populist renaissance." Those two words are an oxymoron. Populism and class warfare politics has never made any country richer, or solved any actual problems in the real world. At its best its tanks economies with nonsensical and unsupported economics "theories" and wishful thinking. See Peronism in Argentina, which a healthy does of Milei libertarianism is fixing finally after decades of stagnation and wild swings from positive to negative GDP.

(GDP in Peron times)

At it's worst extremes it's resulted in lots and lots of people dying:

The Great Leap Forward was supposed to make China an industrial powerhouse and instead is destroyed agriculture and saw roughly 50 million Chinese die from famine and starvation. The Four Pests Campaign saw Chinese people attacking the "four pests" that killed crops, including Sparrows. Chinese peasants banged drums and pots and made loud noises so the birds couldn't land and died of exhaustion. When the actual parasites started eating the crops, there were no sparrows to eat them and the wave of mass crop failures accelerate the starvation of millions. Horrible ideas lead to horrible results like beads on a string. One naturally follows the other if these ideas are allowed to play out in reality.

(Source: A Chinese propaganda poster from the Mao era, showing the exact opposite of how it turned out with the Great Leap Forward)

China only became the modern marvel China of today after the death of Mao and the actual renaissance of Deng Xiaoping's Reform and Opening Up market reforms. In short, they embraced capitalism and markets and ended class warfare. Today China is second largest economy in the world and on its way to becoming the largest and they did it by doing a 180 and building.

The chart below tells the entire story. See if you can spot the moment when China started to make market reforms.

(Source: the Cato institute)

We're not solving any climate problems because you turned off your LED lamp for a half hour more or decided not to shower every day. In that case, you've successfully increased the discomfort of your spouse and co-workers, but you definitely have not prevented a climate apocalypse by raising your stink level and reading in the dark.

Big climate shifts will be fixed by technology, by a boom in clean energy, that's already underfoot with wind, solar and geothermal but needs to embrace nuclear with open arms.

You also don't get there by making society poorer.

Poor societies don't spend any time talking about how they can save the environment or raise living standards or "pause AI" because they can't. When you're stuck at the bottom of Maslow's Hierarchy of Needs and just trying to put food on the table and find clean water, you don't get to worry about fantasies of AI superintelligence run amok.

The truth is more money does make us happier.

When you get to choose where you work instead of taking the only jobs available, that's a better life for you and your family. When you get to argue at the dinner table about politics, it's because you have food on the table in the first place. When you get to complain on the internet about rabid consumerism and the evils of technology, you have that choice because of the datacenters that power your choice to complain and the existence of the very social media you despise letting you post about how much you hate social media.

In a fantastic scene in the Aviator, Howard Hughes loses patience listening to a rich family saying "oh we don't care about money" and says "that's because you've always had it." When you've lived in a rich society that your ancestors created and known nothing but peace because your wars are fought on TV by remote control and drone strikes, you get to complain about "rabid industrialism" and "super consumers" and other made up nonsense.

When we have a rapidly growing economy the benefits ripple across society. We get not only positive first order effects but second order effects too. People are richer, healthier, they've got more opportunities to improve their lot and their children's lot in life, and then politics and trade become much more stable.

The Killingsworth study out of the University of Pennsylvania in 2021, "found that happiness rose steadily with income well beyond $75,000, without evidence of a plateau."

“In the simplest terms, this suggests that for most people larger incomes are associated with greater happiness,” says Killingsworth, a senior fellow at Penn’s Wharton School and lead paper author. “The exception is people who are financially well-off but unhappy. For instance, if you’re rich and miserable, more money won’t help. For everyone else, more money was associated with higher happiness to somewhat varying degrees.”

In other words if you're already a miserable person who's constantly unhappy anyway, more money won't make you happier. But for the rest of us, a growing economy, and economic dynamism are the keys to a longer, happier, healthier life.

Money money means more happiness for a few simple reasons. One of the best is that it's a kind of frozen possibility that buys you options.

Take this simple example: A moderately upper middle class friend told me that while hiking on the Camino de Santiago recently, he was exhausted after 10 days and decided to spring for a single room, instead of a room with 4 to 10 people, as many pilgrims do to save money on the trip. That left him refreshed and ready to go again the next day as he'd finally recovered from the punishing toll the hike took on his body. But fellow pilgrims he met along the way weren't doing so well after a similar amount of days. They were sleeping in rooms with lots of people snoring and making noise and so their sleep was light or non-existent and that meant their bodies didn't have a chance to properly recover. One of them fell and broke her leg because she was so exhausted. Money bought him the possibility of better sleep and better sleep translated into a better experience the next day.

It's seems money can buy happiness or as Janis Joplin once said, "money can't buy happiness but it can buy a big old pink Cadillac to drive around looking for it."

In the book How Economics Explains the World, author Andrew Leigh writes about what it means for society to get better at producing more with less for everyone:

"In prehistoric times, the only source of artificial light was a wood fire. To produce as much light as a regular household lightbulb now gives off in an hour would have taken our prehistoric ancestors fifty-eight hours of foraging for timber.

"By Babylonian times, the best lighting technology was a lamp that burned sesame oil. To produce the same amount of light, a Babylonian worker in around 1750 BCE would have had to work for forty-one hours. An earthenware lamp, which used a cotton wick and oil or ghee.

"Then came candles. Initially made from animal fat, they were time-consuming to produce (and smelt awful). Even in the late 1700s, the typical worker would have to devote five hours’ work to producing candles that emitted as much light as a regular household lightbulb gives off in an hour.

"Through the 1800s came developments in gas lamps, which reduced the time cost of an hour of light to a few hours of work. With the invention of the electric bulb, light got cheaper still. By the early 1900s, it took just minutes of work to buy an hour of light.

"Today, less than one second of work will earn you enough money to run a modern household lightbulb for an hour.

"Measured in terms of artificial light, the earnings from work are 300,000 times higher today than they were in prehistoric times, and 30,000 times higher than they were in 1800. Where our ancient ancestors once toiled to brighten their nights, we rarely even think about the cost as we flick on a light."

Or as James Pethokoukis writes: "The benefits of economic growth, driven by technological progress, are myriad. Growth raises living standards, makes us healthier, and creates a more interesting world of greater fairness, tolerance, and opportunity. Even during the ongoing Great Downshift, Americans are better off than they used to be across all those metrics...And globally, the decline in extreme poverty has been nothing short of miraculous thanks to faster growth enabled by tech progress—including global communication and the standardized shipping container—especially in Asia. Yet as important as all those things are, and as important as future advances might be, we shouldn’t ignore the most important benefit of all: an Up Wing society is one capable of solving big problems, even if those problems are unforeseen or poorly prepared for. An Up Wing society is a more resilient society."

What does he mean by a resilient society? A society with extra wealth and resources is a society that can turn on a dime and adapt to rapid and evolving threats quickly. A poor society that's just trying to figure out how to feed its population and keep bandits from robbing people in the street has no energy and resources left over to join together to fight something bigger.

Look at the COVID pandemic. America started off incredibly ill prepared for the pandemic. Article after article told us how we'd become beggars in the richest society on Earth, begging for more medicines, more masks and more ventilators. Old anti-virus technology took years, if not decades to produce vaccines. mRNA was a moonshot technology, still in its infancy and it was Operation Warp Speed, under president Trump, where it accelerated to incredible new heights. We developed, tested and delivered a vaccine at scale in under year, achieving 40% coverage faster by an order of magnitude than during any other outbreak in history.

The fantastic movie Contagion proved terrifyingly accurate about lockdowns, panic buying and the rapid and wild spread of misinformation and anti-science influencers.

But what it failed to predict was progress.

If COVID had happened during the time the movie came out, in 2011, only a decade earlier, it would've proved accurate on the time it took to come up with a vaccine through trial and error because Moderna was just a twinkle in its founders eye. But because medical technology had advanced we were able to prevent a massive amount of serious sickness and death. The National Library of Medicine puts the numbers in stark relief here:

"Based on reported COVID‐19 deaths, vaccinations prevented an estimated 14.4 million deaths (95% credible interval [Crl] 13.7–15.9) from COVID‐19 in a year. However, if excess deaths were used, this estimate rose to 19.8 million (95% Crl 19.1–20.4) deaths prevented, equating to a global reduction of 63% in total deaths (19.8 million of 31.4 million) during the first year of COVID‐19 vaccination."

Progress is what degrowthers and doomsayers get wrong year after year after year, no matter if their doomsday is caused by a new ice age, a population explosion, rabid consumerism or AI. They miss the second order beneficial effects of stronger growth and new innovation and technology.

In my article the Hummingbird and the Poison Tree, and my upcoming book of the same name, I wrote about the characteristics that mark a healthy, growing system and unhealthy, decaying systems. Remarkably they're largely the same characteristics whether we're talking about a system of two, a marriage or a friendship, a larger system like a company, or a mega-system like a nation state.

The short version is that the Hummingbird Effect is a positive reinforcement loop of beneficial evolution and a rise in second order positive effects. Better economics and living standards lead to a positive rise in happiness, technological innovation and artistic expression. In others words, good begets good.

The Poison Tree Effect is its dark doppelganger.

It came out of phrase I coined years ago: "Poison trees make poison fruit," or just "Poison tree, poison fruit." Sometimes a system is resilient enough to have poison flowers and poison trees in the ecosystem. There are enough beneficial plants to keep the parasitic plants at the fringes, doing light damage but never taking down the entire system.

But sometimes those poison trees spread, choking out the rest of the forest, their seeds floating away on the wind and infecting distant ecosystems and their fruit killing off the animals and insects that eat it and causing a chain reaction of decay. Too many poison trees and a system collapses.

Both of these ideas link up neatly with another idea that comes from Malcolm Gladwell: The Tipping Point.

That's a point when ideas and events in a system hit critical mass the system surges up or down dramatically in a way that's virtually unstoppable. Once the printing press takes off, its power ripples out into the world and society undergoes tremendously positive changes, with literacy rising, knowledge spreading and advances in science and medicine taking hold, while the darkness of the middle ages starts to recede.

If you focus on degrowth, class warfare, protectionism, immigrants "stealing all the jobs", you create a lethal combination of almost inevitable failure for your future.

On the whole, if there are more positive characteristics, then a system is much more likely to keep trending in the right direction and you can ride the tide to prosperity and joy. Things become a self-fulfilling positive prophecy.

The Hummingbird and the Poison Tree and the Tipping Point don't just apply to the evolution of technology but to the evolution of ideas. They point to the deep interconnectedness of everything from ideas, to attitudes, to balances and imbalances in a society. A society that doesn't invest in the future and doesn't believe there can be a better tomorrow digs its own grave and ensures that's exactly what happens.

But how do we get to a world with an explosion of positive first and second order effects and avoid a downward spiral?

Getting Back to the Future

The key to a better tomorrow is laying the groundwork for a society that flourishes and grows is to invest in the long term.

When we have the right policies and foundation in place, we don't know exactly what will come out of the magic box of ideas and innovation, we just know it will have wildly beneficial effects for a huge swath of people across society. Not every policy or technology will work for everyone, in every way, as that's impossible, but the more good policies we have in place the more we see a broadly beneficial overlap of prosperity across class, race and creed.

To have a healthy, growing society, you want to foster the characteristics of a hummingbird society with a strong foundational framework of policies and ideas.

That framework is what gives us the best chance of a brighter future.

So let's look at seven policies that can help us accelerate America and turn it into a bright, shining light for the world.

1. DARPA and the Operation Warp Speed model everywhere.

Let's start with start with a DARPA style initiatives for everything. Energy. Agriculture. Nanotech. AI. Biosciences.

To create the future you have to invest in the future at the public and private level.

While many people today look at government as an inefficient mess and business as the gold standard for innovation and leadership, that's only half the story. It's true that business is more dynamic and makes industries more scalable and repeatable but it's often government that's been the early driver of ideas when they're too risky or too early to justify the cost of R&D. Think of NASA and the moon landing, the Human Genome Project and DARPA, which helped usher in technologies like "Moderna's COVID-19 vaccine ... weather satellites, GPS, drones, stealth technology, voice interfaces, the personal computer and the internet," writes the Economist.

Business can rarely spend the time or money on moonshot projects unless they're already an effective monopoly like AT&T and the 1950s, which spawned Bell Labs and its long term thinking, or Google and DeepMind in the modern era.

In other words, you need to have a massive war chest and an insanely mature and profitable business somewhere else to spend on such long term vision. That's often where the government has stepped in over the years. The vast majority of federal funding, which is now badly under siege, is used for basic research and discovery while R&D in the private sector focuses on making things smaller, better, more repeatable and scaling the commercialization of existing, working products. Both government and business are essential pieces of the puzzle.

As Pethokoukis writes in the Conservative Futurist: "At the height of the 1960s Space Race, the United States spent nearly 3 percent of its economy on scientific and technological research and development. Of that total, about 2 percent was government funding and 1 percent business funding. Today, the United States spends just over 3 percent on scientific and technological R&D with a bit more than 2 percent conducted by business and about 0.75 percent by government. (All the various spending bills passed during the first half of the Biden administration—the Bipartisan Infrastructure Act, the CHIPS and Science Act, the Inflation Reduction Act—didn’t change federal R&D investment that much.)"

In other words, the government has dropped the ball and isn't spending enough and it hasn't been for a long time. The latest administration’s attacks on universities and foreign researchers threatens to make it even worse.

The only recent example of rapid spending and progress at the government level was Operation Warp Speed, one of the most incredibly successful technological acceleration projects in history, that has the dubious distinction of being largely disavowed by both the left and the right, the left because it came under Trump's first administration and the right because of the base's sudden aversion to vaccines. But Operation Warp Speed was massively successful. It saw the US deliver a vaccine in 9 months and reach 40% coverage in the first year, instead of a decade.

As Klein and Thompson write in Abundance: "Operation Warp Speed is the oddest political orphan. A program named after Star Trek has disappeared into its own kind of black hole. A policy that stimulated the economy more than the Apollo program, and which may have saved more lives than the Manhattan Project, has almost no loud champions in politics. Even its scarce champions seem intent on taking the wrong lessons from its success.

"In an essay for the Wall Street Journal, the University of Chicago professor Casey B. Mulligan, who had served as chief economist for the Trump White House, claimed that 'the urgent lesson' from Operation Warp Speed was that 'too much government hinders private innovation.'"

"The right lesson from World War II and Warp Speed is that the state is no enemy of invention or innovation. In fact, the government can accelerate both. In the 1940s, the Office of Scientific and Research Development mapped out the chemistry and production challenges for penicillin and turned an obstacle course into a glide path. In 2020, the US government similarly identified the bottlenecks to rapid vaccine development and removed them. In both cases, the government served as a chief national problem solver, molding its policies to fit the moment. It is a vision of a new kind of entrepreneurial state. It is the government as a bottleneck detective."

Or as Pethokoukis writes in the Conservative Futurist: "America was the first nation to roll out a vaccine. And it did so rapidly, getting the COVID-19 vaccine to the American people in less than ten months. And it wasn’t just a single vaccine. Operation Warp Speed delivered three in record time. What’s more, there’s something in its success for everybody. It was an acknowledgment that drug companies had lost money during previous public health emergencies, such as the 2014 Ebola outbreaks, when treatments they developed turned out to be unnecessary. But now we needed those companies to make numerous big and bold—and risky—bets. So Operation Warp Speed placed orders for vaccines and therapies while still undergoing clinical trials, regardless of the outcomes. This encouraged pharmaceutical companies to expand manufacturing capacity so vaccines and therapies were ready to be distributed once they had the FDA’s green light. There you had it: the private sector innovating with the public sector providing a safety net. The success of Operation Warp Speed also showed the overall strength of the American Innovation System."

We need more DARPA and more Warp Speed, more sci-fi inspired thinking and we need it now.

DARPA is one of the most successful government agencies of all time. It's success comes from its unique approach to funding and managing research, focusing on high-risk, high-reward projects and fostering a culture of innovation. It's flat structure, rapid project turnover, flexible contracting, and the ability to attract and retain top talent through a unique combination of autonomy and accountability, mirrors the best of Silicon Valley on the 1990s internet innovators.

DARPA's small, flat organizational structure, with frequent program manager and director changes, enables rapid idea generation and adaptation. The short project durations (typically 3-5 years) and lack of renewals encourage a focus on delivering tangible results within a defined timeframe. It also has the flexibility to issue various types of contracts and grants and hire talent fast, often at higher salaries than typical government employees, to attract and retain top-tier expertise. It also prioritizes projects that push the boundaries of science and technology, even if the chances of success aren't guaranteed. That means when those technologies work the breakthrough is massive and beyond anything that might come out of the paper pushing and risk averse research community.

We need to replicate this approach to all aspects of the economy, everything from semiconductors, to energy, to agriculture, to nanotech, to travel and communication. Everything. Wherever a DARPA model can be used it should be used right now.

T o build the future first you have to do the basic research that provides the building blocks for the innovators of tomorrow.

We don't know what that breakthrough will be that provides the catalyst to some starry eyed innovator years later, but if we don't do the research in the first place that leads to glasses, then we never get the microscope, and if we don't get the microscope we don't get germ theory and we don't get modern medicine.

Everything follows everything else in a great chain of progress.

2. Slash Regulation and Build Stuff. Make regulation targeted and streamlined. Put less people in the way.

On the left and the right people are waking up to the glut of zoning regulations preventing people from building new housing and apartments to the absurdly long wait times to get approval to build anything, whether that is a bullet train or a new nuclear energy stations.

As Nolan Gray writes in the Atlantic: "Until recently, zoning was a sleepy backwater in the policy world...If a layperson knew anything about zoning, chances are she didn’t have an opinion about it...

"A decade of urban upheaval has changed all of that. Amid ongoing crises of housing affordability, inequality, segregation, and sprawl, fixing zoning has emerged as a cause célèbre. Over the past decade, the Obama, Trump, and Biden administrations have all vowed to take on zoning reform. National outlets now regularly decry the evils of zoning. And local YIMBY groups across the country are rewriting old ordinances. For the first time in 100 years, the arbitrary lines that divide up the American city have become impossible for the average person to ignore.

"In 2021 alone, home prices rose by nearly 20 percent. Over the same period, rents also skyrocketed, prompting fears of a surge in homelessness. Once contained to the coasts, the crisis is spreading inland; home prices in previously affordable cities have hit record highs...Boring though it may be, zoning has real-world effects that are dire. Between mandating parking garages and banning apartments, it has made infill development prohibitively difficult in many American cities. And in suburbs across the country, it has made the starter homes we so desperately need—think townhouses and homes on small lots—effectively illegal to build."

This isn't just a problem in America but all over the world. We're not building enough homes. This results in ridiculous policies like we see in Spain, where the Spanish government attacks AirBnB and tells it to yank 65,000 listings while the president vows a 100% tax on foreign buyers. None of that builds a single new home for anyone. Landlords simply skirt the rules and rent it to the highest bidder.

But sound, sane policies rarely seem to get the right traction. Instead, insane, ineffective policies abound around the world and in America.

There's only one way to make prices go down:

Increase supply to meet demand. Economics 101.

Only building will fix the problem. If tomorrow, Spain passed a law that let every building owner add one new floor in its major cities and streamlined the permits and gave builders a ten year tax break on the build out, the housing problem would be solved virtually overnight. If America slashed the red tape and the impact studies and the constant approval and permits to do anything, the problem would fix itself.

We've got to clear the way and make it affordable to build again so that builders don't have to file plans three times, see their costs skyrocket and their times to build on the land the own slowed to a crawl.

Or as Pethokoukis writes in the Conservative Futurist: "Economists Chang-Tai Hsieh and Enrico Moretti find that the “creeping web of these regulations has smothered wage and gross domestic product growth in American cities by a stunning 50 percent over the past 50 years.” Let’s say just three high-productivity cities—New York, San Jose, and San Francisco—changed their housing rules to the level of restrictiveness of the median U.S. city. The net effect of this increased density would be a 4 percent increase in U.S. GDP and an additional $4,000 in average earnings for all American workers. Regulation costs in home building are 11 percent higher today than five years ago, accounting for $94,000 of the final price of the average new single-family home. Over the past decade, regulation costs have increased by 44 percent."

But it's not just zoning laws. The 1970s National Environmental Policy Act had a lot of good ideas but a lot of bad second order effects. Everyone loves clean air and clean water but there's now little doubt that this tiny piece of legislation has made it incredibly cost prohibitive to build just about anything that we really need from energy infrastructure to bullet trains. The chances of repealing NEPA are basically zero but we should be looking to streamline its impact which has reached absurd proportions.

On average, the Conservative Coalition for Climate Improvements, found that NEPA's environmental review process adds about $4.2 million to complete a single review and takes 4.5 years. Analysis from the American Action Forum found that "148 energy and transit projects in NEPA review were estimated to cost $229 billion." Some estimates suggest that federal regulations, including those related to NEPA, can balloon a new energy or infrastructure project's by up to 20%.

Right now, China's entrepreneurs and its government is investing $590 billion yuan (81B USD) in new trains, laid 48,000 kilometers of bullet train tracks, see 4.08 billion passengers ride the rails, and moved 3.99 metric tons of cargo.

Meanwhile in California, a single bullet train is expected to cost $88-$129 billion dollars and environmental laws have made building a single bullet train nearly impossible. The project has to comply with both state (CEQA) and federal (NEPA) environmental regulations, requiring thorough environmental assessments and mitigation measures. It's already undergone extensive environmental reviews to assess potential impacts on air quality, water resources, wildlife habitats, and other environmental factors. These reviews are documented in Environmental Impact Report/Environmental Impact Statement (EIR/EIS) documents, which are released for public review and comment. Opponents of the project have filed numerous lawsuits, challenging the environmental clearances and claiming violations of environmental laws. Some of these lawsuits have been successful in delaying the project or requiring more environmental reviews and wrapping it up in a nasty tangle of red tape.

The only way to fix this is to strip back these laws and streamline them. We've got to limit the lawfare industry that makes money suing the state to keep trains locked up in a massive kafkaesque nightmare of legal battles.

To do that we need a more practical and modern environmental movement, one that's less about singing songs and protesting nuclear and more about more nuclear and more targeted legislation in the places that matter and less regulation in the places where laws screw everything up.

We're already starting to see a more pragmatic and practical environmental movement. California is already reworking its CEQA bill to let people build more easily in a win for YIMBY advocates. The bill has been one of the biggest barriers in the blue state to building anything from trains to housing for years.

We don't need an environmental movement focused on degrowth, turning of lights and singing songs on acid in the desert, we need on that sees technology like batteries, nuclear and solar as the wave of the future. Technological solutions like better energy and less regulation that speeds technological progress and gives us the tools we need to make a greener and more prosperous tomorrow. Climate change is a clean energy problem not a problem of rabid super consumers and meat eaters.

More solar. More nuclear. More wind. More batteries. Not just some of it. All of it. Faster.

We beat the climate crisis with big technology not wishful thinking.

3. More immigration, not less. More trade, not less.

Let's turn to trade now. We've seen a massive surge in Peron style protectionism and protectionism has never led to anything good in any country on Earth.

Trade barriers do nothing but make things more expensive for people right here at home. They're taxes. Other countries do not pay them. We do. You and me. Our small and big businesses. We pay them. Simple as that, simple math. There is no logic that says making things more expensive for people makes them richer or better off in any way whatsoever. We do not make life more cost effective for working Americans by making it more expensive to buy iPhones and groceries.

In the first Trump administration we saw tariffs as high as 19.3% on goods from China, which were kept in place by the Biden administration, and now they've rocketed to 30% and they're just as high elsewhere, leading to a crisis and dramatic drops in shipments and suffering small businesses.

This is a disaster in the making, threatening to destroy the more powerful economy in the world with a self inflicted gunshot wound. We need to go in the opposite direction and get back to basics.

The Peterson Institute for International Economics found that the payoff to America for world trade from 1950 to 2022 was 2.6 trillion USD and lifted GDP by 10%.

They write "the reintroduction of the Smoot-Hawley tariff regime causes GDP to decline by 2.4 percent. The introduction of reciprocal tariff regimes by US trading partners further contracts the US economy by 2.1 percent. In all, the Bradford and Lawrence simulations indicate that the US economy is 4.5 percent larger due to sharply lower worldwide tariff barriers since 1947. This equates to $500 billion in 2003. The calculated decrease of $500 billion in US GDP illustrates the impact of a shock back to 1930s-style protectionism rather than a parallel universe without liberalization over the past 50 years."

They found that the elimination of all barriers "would raise US GDP by about 4.1 percent. Scaled to 2003, this amounts to an additional $450 billion in US GDP, $1,500 per capita, or $4,000 income increase per household annually."

Those numbers would be significantly higher today around $785 billion in USD in additional revenue. It does not work the other way and it never will, even with magical thinking.

Even worse than keeping foreign goods out of the country is keeping skilled people out.

Making it so the best and brightest don't want to come here and study and work anymore is a disaster that makes the California bullet train look like a success by contrast. When we don't have a bullet train we can still fly. But when we don't have the best and brightest coming here that means they don't start new businesses, or staff the R&D departments of our biggest and best companies.

An MIT study found that immigrants were about 80% more likely to start a company and that "'Immigrants found more firms in every bucket...They create more firms, they create more small firms, they create more medium-size firms, they create more large firms.” He adds: “It’s not the case that [immigrants] only create growth-oriented startups. It’s not the case they just create subsistence businesses. They create all kinds of businesses, and they create a lot of them.”

Julia Yoon in an article for the Center for Strategic and International Studies, writes that "According to the National Science Foundation (NSF)’s 2024 Indicators Report on the state of U.S. science and engineering (S&E), foreign-born workers comprise approximately 19 percent of the overall U.S. STEM workforce in 2021, inclusive of citizens and non-citizens. However, among the U.S.’s most highly educated STEM workforce cohort, foreign-born representation is dramatically higher, with nearly 60 percent of doctorate-level computer and mathematical scientists (58 percent) and doctorate-level engineers employed across all S&E fields (56 percent) in the U.S. are foreign-born."

48% of all Fortune 500 companies have an immigrant founder.

When we see America at the Olympics winning metal after metal, take a close look at the people standing on that podium and you'll see a lot people who came here for a better a life. Why can't any other country win as many metals, even China with a population that's 4.3 times the size? When you're the place that people want to send their children for a better life, you've got a massive advantage in everything from sports to the sciences to business because the best of the best come to play and work for your team.

Highly skilled education is a super power and one we can't take lightly or set on fire so that nobody wants to come here to work. When we send home a promising machine learning researcher over a fishing license violation and speeding tickets we're squandering that superpower and making it more likely that tomorrow's brilliant kids go somewhere else to build a better tomorrow.

Everyone here, other than Native Americans, was the son or daughter of an immigrant at some point in their history. We're a nation built from immigration and people who came across the seas to build a better life. We'd do well to remember it going into the next century or we'll wind up as a superpower of the past, instead of a superpower of the future.

4. School. Trades. Cooking. Shop. Engineering. Science that's fun. Critical Thinking.

Every teacher I talk to talks about teaching to the test. Common core is a mess. The joy of school is gone. We don't do real science experiments in schools, we read about them. We memorize facts instead of teaching kids to think. The students hate it. The teachers hate it.

Why are we still doing it?

Who the hell knows.

Sunken cost fallacy?

It seemed like a good idea at the time, as we tried to emulate Asia and get more kids doing well on tests. They're doing worse and that's because the joy is gone.

Everyone remembers that one great teacher they had that changed their life, that made them care about what they were learning, that made them connect the knowledge to the real world. We need more of those teachers and more of the kinds of freedom they had to make education great for everyone who had the good fortune to pass through their classroom.

At a recent family reunion, I met a teacher friend of one of my cousins. He told me with pride how he used to have the kids build Rube Goldberg machines as an experiment in thinking and creativity and how years later he still got letters or praise from kids who'd gone on to be engineers or scientists. He'd run into a former student and he/she'd say "are you still doing that Rube Goldberg machine?" and he'd shake his head sadly and say no, the school squashed it.

Hands on classes are the future. Learn by doing, the old fashioned way works. Science is where experiments actually happen instead of just reading about it in a dry text book. Get the phones out of the classroom. Kids can browse TikTok and message their friends later. Ditch common core. Every single teacher I've ever talked to about it, hates it as utter nonsense that did nothing to make kids brighter or more adaptable.

But it's not just about science, we've forgotten the real world in school.

My grandmother learned how to balance a checkbook and do her taxes in school. She learned how to cook. My grandfather had shop and sports. We've forgotten the hands-on lessons of life. If you want to be healthy, you better learn to cook, or you'll be eating 10X the salt and fat in restaurants. Is it any wonder so many people are obese?

If you want to understand math, make that math practical. If you want to know how to fix things in the real world, you've got to get your hands dirty. We need blue collar workers as much as white collar workers. Not everyone is meant to start a company or work in an office. If kids never learn to cook or build things then we miss out on many of the kinds of skills we need for the jobs of tomorrow or just good life skills like making your own healthy meals.

And teach kids about the people that matter. Ditch dry, boring, horrible text books that are subject to lobbying and political activism. Teach kids about how the tea kettle got made and the skyscraper, or who discovered DNA and who built the microscope and less about Hitler and politicians down through the ages.

While we're at it, give every kid a personal AI right out of the The Diamond Age. Short bursts of individual tutoring make a big difference, but we can't scale it right now. AI will change all that. It’s already quietly impacting schools and the smartest kids and teachers already embrace them instead of fearing and fighting it.

More than anything we need schools that teach kids to think. Many of our problems today come from the fact that people who made those policies and decisions never learned to think clearly. If you've never learned to think you're bound to come up with stupid ideas. My greatest teacher in school taught me that critical thinking is the most essential skill in all of life and it's never failed me, allowing me to adapt to changing economic environments, learning new jobs on the fly by trying and failing and getting back up and trying again, and just by figuring things out on my own.

Critical thinking. The arts. Music. Everything matters. A great teacher once said "education is what you're exposed to." Today we expose kids to a boring slop of memorized facts and we're ripped the learning out of education, the excitement, the joy and the fun.

Of course, not all learning is fun. Not all learning is excitement. But when you have none of it, you've set back an entire generation of people won't be able to power next wave of American greatness.

5. Manufacturing

You learn by doing and that's what the Chinese have been doing with manufacturing since the 1980s after the 1990s.

People don't go to China because it's cheap. They go because they have the most advanced manufacturing in the world. China long ago stopped being the cheap place to build stuff. That's Vietnam and other places now.

Tim Cook, CEO of Apple, said that "there's a confusion about China...the popular conception is that companies come to China because of low labor costs. I'm not sure what part of China they go to but the truth is that China stopped being the low labor cost country many years ago and that is not the reason to come to China from a supply point of view. The reason is because of the skill and the quantity of skill in one location and the type of skill it is...the products that we build require really advanced tooling and the precision that you have to have in tooling and working with the materials that we do are state of the art. And the tooling skill is very deep here. In the US you could have a meeting of tool engineers and I'm not sure we could fill the room. In China you could fill multiple football fields."

We are not going to bring back American manufacturing with tariffs and the kinds of manufacturing that we need is not the manufacturing that we saw in the 1950s. The Chinese saw that as a dead end and wisely invested in robotics. We haven't.

As I wrote in my article on Why We Need An American DeepSeek:

In a brutal wake up call to the west called America is Missing the New Labor Economy, SemiAnalysis shows that America and the EU are badly behind on roboticization and the race to general purpose robotics.

When it comes to the sheer number of robots deployed in Chinese factories, nobody else in the world is even close. Take a look at that chart below. Even an idiot can see who is winning that race.

(Source: SemiAnalysis)

The authors paint a dire picture for the west:

"This is a Call for Action for the United States of America and the West. We are in the early precipice of a nonlinear transformation in industrial society, but the bedrock the US is standing on is shaky. Automation and robotics is currently undergoing a revolution that will enable full-scale automation of all manufacturing and mission-critical industries. These intelligent robotics systems will be the first ever additional industrial piece that is not supplemental but fully additive - 24/7 labor with higher throughput than any human—, allowing for massive expansion in production capacities past adding another human unit of work.

"The only country that is positioned to capture this level of automation is currently China, and should China achieve it without the US following suit, the production expansion will be granted only to China, posing an existential threat to the US as it is outcompeted in all capacities.

Robotics is about to explode across the economy and make the past industrial revolutions look tiny by comparison. Picture factories churning out robots that build still more robots. Each iteration cuts costs, boosts quality, and powers a production flywheel that keeps spinning ad infinitum—leaving rivals more and more desperate and breathless as they sprint to catch up but find out its hopeless.

Because robotics is a true general-purpose technology, it will upend everything from textiles and electronics to consumer goods.

Meanwhile, the West is about to get caught flat footed as the unexpected punch smashes into our face. Here's a fully automated factory in China cranking out the components for a thousand missiles a day 24/7. Meanwhile American factories are still producing missiles like its 1945, by throwing people at it. A Pentagon war simulation said the US runs out of missiles "in days" in a war with China because it can't match China's production capacity. Europe’s industrial core is getting devoured by China’s relentless rise and its absolute inability to source enough power (while shutting down their nuclear reactors), and the US keeps fixating on other markets while banking on cheap offshore production.

To fix American manufacturing we've got to think in ten year plans.

If we want to rebuild our manufacturing we need wise, ten year thinking. Ten year 5% tax rate for all factory builders foreign and domestic. Cheap loans. Streamlined exceptions to environmental impact studies and streamlined permits. The Asian countries do this best. Right now the country most poised to dominate semiconductor manufacturing after Taiwan, is South Korea, not the US.

Samsung currently operates four advanced semiconductor fabs in South Korea focused on cutting-edge nodes (4nm, 3nm, and 2nm), with additional facilities under construction. They are behind in total capacity versus TSMC for now. TSMC retains a significant lead in total wafer production capacity, but Samsung is aggressively expanding to close the gap with a combination of national and private funding. Samsung is gunning to be the top manufacturer in the world and they're working on the Yongin Mega Cluster (1) (2), six fabs capable of 2nm and below, part of a $471 billion national project with Samsung and other firms, first phase operational by 2030. The goal is 7.7 million wafers per month starting in 2030.

It's important to understand that chip production is not really a pure private endeavor. For mega-scale chip manufacturing to work, it's always been a potent combination of government and private industry. Governments often invest massively, clear regulatory hurdles and give sweetheart tax breaks to own this lucrative industry. This is true in both South Korea and Taiwan and throughout Asia, a style of business that's not natural to US industry outside of the 1950s when government was more tightly in step with business after the war.

Take TSMC and its origin story for instance: "Alongside generous tax benefits, the Taiwanese government, through the National Development Fund, Executive Yuan, provided another 48 percent of the startup capital for TSMC, and the rest of the capital was raised from several of the island's wealthiest families, who owned firms that specialized in plastics, textiles, and chemicals. These wealthy Taiwanese were directly 'asked' by the government to invest. 'What generally happened was that one of the ministers in the government would call a businessman in Taiwan,' Chang explained, 'to get him to invest.' From day one, TSMC was not really a private business: it was a project of the Taiwanese state."

There's more. When "K.T. Li (李國鼎), known as the “father of Taiwan’s economic miracle,” convinced Taiwanese-American Morris Chang (張忠謀) to move back to Taiwan and help build up Taiwan’s semiconductor industry. When Chang founded TSMC in 1987, he did so with $100 million in seed money from the Taiwanese government."(1)

TSMC benefits from massive tax credits and multi-year tax reduction deals. The same is true in Korea, where the new chip clusters will benefit from cleared hurdles for power plants and tax benefits for R&D and more. Expect the effective tax rate of the facilities to be roughly 0% for many years when all the numbers are actually crunched and all the credits and building costs are applied.

We need to start doing the same here. Prioritize. Incentivize. Clear roadblocks and take a ten year view of these technologies, not a ten second view. Rip it all right from the pages of the TSMC playbook where the government clears roadblocks, lowers taxes for the chip giants and greenlights new energy sources (History of TSMC).

To win, you have to think bigger and better, but you have to do it wisely.

6. Visions of a Positive Future. New school Sci-Fi.

The cyberpunk aesthetic has dominated sci-fi since the 1980s. It's not always direct but its tendrils of influence extend into everything, even outside of sci-fi. The dark, rainy future is the backdrop of a million movies and TV shows.

(Source: Bladerunner)

It's old. It's tired.

We need something new, a more positive future, a more nuanced one.

People need to see science and tech solving big problems and making lives better. It's a cheap trick to make the robot go rogue, like Ultron in the Avengers. It's nothing but the big bad with a metal skin slapped on. It's easy. It requires zero thought and no understanding of technology whatsoever.

We used to have very different futures. The Star Trek universe saw us as a multiplanetary species with replicators that could print up anything we wanted and where we didn't even need money anymore because we had so much abundance. The Jetson's saw us jetting around in flying cars and working a few hours a week. In a famous scene George Jetson comes home exhausted because he had to "push the button on and off five times" at work.

That was a different time and a different way of looking at technology and the world.

It changed because our relationship to technology and society changed. The upbeat, pro-tech, abundance sci-fi of the 50s and 60s gave way to the rainy, neon soaked landscapes of the cyberpunks, where everything's falling apart and evil corporations dominate our lives.

Part of that was the rise of big business in the 1980s and just a natural, artistic reaction to golden, glowing futures. The gritty, drizzly, dark noir futures of Terminator and Bladerunner swung the pendulum in the other direction.

But it's gone too far.

Too many people have grown up with nothing but Black Mirror and Termintor and can't remember a time where technology might lead us to a future free of disease and material worries, where mRNA treatments beat cancer and self-driving cars save millions of lives because their drivers don't get distracted.

Don't get me wrong. I love a good dystopian story just like anyone. I fell in love with books like Neuromancer in college. It's gritty, streetwise world of drug dealers, displaced augmented soldiers, lowlifes, punks and hackers was a powerful antidote to the glittering and glowing golden age of sci-fi where tech was treated as if never had downside or a flaw (everything does). I sunk hundreds of hours into Cyberpunk 2077. I've read every cyberpunk classic from the early 80s and gorged on anime masterpieces like Ghost in the Shell. My first novel, the Scorpion Game, is dripping with dystopian biopunk organic cityscapes and slashing rain. I love the look and feel of a noir future and so does everyone else apparently, because it's become the tired trope of almost every movie and video game of the past forty years.

And it often gets the future right. Below is one of the latest covers of nueromancer and on the right is a military drone operator in the Ukraine war.

But it’s only a partial slice of reality. There's something badly missing today, where we have an incredible advancement like AI predicting billions of protein folds, revolutionizing medicine, and winning its creators the Noble prize but it barely gets a mention in the news while even totally mundane and inconsequential news at OpenAI gets framed as some kind of harbinger of the rise of Terminators.

What we desperately need now is nuance.

It's easy to paint social media as an all pervasive evil in articles and shows like Black Mirror. It's a thousand times harder to paint some of its downsides in a realistic way but also show that social media has allowed billions of people to connect in new ways, make new friends, find love and share ideas. It exposes us to a wider range of viewpoints as much as it narrows our viewpoints. Both are true at the same time. I've met multiple good friends online, met the love of my life there and my story is not unique.

Nuance is key and so is showing how tech and progress solve problems.

The great detective mystery of John Snow, hunting down the source of cholera, a dreaded scourge of a disease for centuries, is a thriller in the making. Using science, math and reasoning, he figured out one of the longest standing and most deadly mysteries in the world. The tiny water pump statue tribute to him in London isn't enough. People need stories where they see how tech and medicine and science transformed lives for the better, where technology is the solution to the evils of the world, as much as it's paradoxically the cause of some ills.

It's already happening. Books like AI 2041, take a more nuanced approach, showing the pluses and minuses of tech but painting tech in big broad, positive strokes. And shows like Murderbot, and the super popular novels that went along with it, invert the AI gone rogue trope and despite its name it's really about a robot who hacks his way free of his "governor module" and just wants to watch streaming TV and generally be left alone.

But we need more. A lot more. We need a new wave of sci-fi and stories that energize the next generation instead of demoralizing them.

7. Electrify the World. Batteries, Bots and Electricity.

Batteries and personal tech are the keys to the next generation of technology. With better batteries, you get renewable fueled grids that can save their power, longer driving cars with incredible range, AR glasses that last all day instead of a few hours, drones and robots and more.

Lithium-ion batteries have already seen a steady march of tiny improvements that have dramatically improved their energy density, while their costs have plummeted. Magnets have gotten stronger and so have transistors. GaN transistors can work at much higher levels of power, like in drones in cars and boats and robots.

But right now we don't really have any major battery champions in America. The top ten are all Asian. China has the biggest, at 35% of the market, followed by Korea and Japan. We can't be content to let the rest of the world make our batteries. They're necessary for everything from our consumer electronics, to the stability of our electric grids, to powering those humanoid bots to work in factories.

In the past, the west led with the last wave of electrification, with incredible pioneers like Tesla, Westinghouse and Edison, but we take it all for granted now. Those days are long gone and today everyone from our cell phone manufacturers to our military buys their batteries from overseas.

Way too often, we're romanticizing the America of the past, with a bunch of people hand-assembling things in factories but those days are gone and rebuilding that won't let us compete with the highly mechanized factories of the future in Asia.

Biden pushed a big infrastructure bill with some strong goodies for electrification but the democrats did the whole country a disservice by linking batteries to the environment alone and forgetting about sovereignty and security. The Ukraine war showed the $500 drones powered by cheap, powerful batteries, are the future of warfare and defense, not just $2 billion dollar stealth airplanes. If you can't make the batteries and source the rare earths at home, you can't build the army of the future. So it's much bigger than a greener planet. Batteries and energy storage are deeply intertwined with reshoring manufacturing, as well as building up our robotics and EV capabilities. The rest of the world wants EVs and they want them in huge numbers and those numbers will only go up and up in the coming decades. We can't simply look backwards and remonantzize the combustion engine and forget about plugging in.

As Noah Smith wrote in a recent blog:

"Whichever country dominates electrical technology will therefore rule the land and the sky in the 21st century. AI is amazing, and yet without drones for it to command, it will not be able to win modern wars. And no matter how entertaining our online lives are, humans will still need to move themselves and their possessions around in physical space, so the auto industry will continue to be incredibly lucrative and important."

To get there we need more the DARPA approach and big incentives from congress and a bigger push in private industry. We need moonshot companies here working on the boring idea of better batteries. The Energizer Bunny is dead and we need to bring him back to life and fast.

It's not just moving about the country in cars and having a wireless PlayStation controller that lasts longer. It's about factories. Defense. Electric grids. AR Glasses. Mixed reality headsets that don't need a cord and a battery pack in your pocket. Drones. Agricultural robots.

Either we start building the electric future now or we watch someone else build it and pass us by.

Accelerated America

If you've made it this far it should be obvious how badly we've stagnated in the west. For too long we've let the forces of deceleration and anti-growth run riot. We got fat and lazy. We've lost our edge.

We forgot that empires are built, not inherited from the great builders of the past.

We've turned to poisonous philosophies like protectionism and populism to try to save us but those cures only make the disease spread faster. We've stagnated so badly that we don't even know how to imagine a better future. That's deadly. A society that can't imagine a better tomorrow can't build a better tomorrow.

If you don't believe you can do something, it becomes a self-fulfilling prophecy. Like the old saying goes before you can do something you've got to believe it's possible. I used to think that was some hippy-dippy, new age nonsense, but as I've gotten older I realize it's literally true. If you don't think you can run a marathon, or climb a mountain, you won't start doing the training to get there and so you'll fail. It's simple math, really.

Instead of accelerating AI so we can have more AlphaFold style medical breakthroughs and self-driving cars, we're talking about stopping datacenter expansion and crippling AI with suffocating AI safety laws built on utter nonsense powered by Hollywood fantasies instead of strong engineering reality.

Instead of expanding science and pushing forward promising new cures for cancer, or HIV, or Zika, we've got populists fighting to ban or limit mRNA vaccines in a classic example of American decel idiocy.

We've strangled our ability to build houses or lay down a bullet train. Our climate laws are broken when they see a bullet train that would take thousands of passengers a day off the roads as an environmental threat.

While China builds hundreds of thousands of miles of railroads, it's become almost impossible to imagine the US building anything like the interstate freeway system that connected the country in the 1950s. Today it would get bogged down by screaming protestors, activists on Twitter, a thousand environmental impact studies and endless waves of permits and approvals and get deluged with a wave of lawsuits.

We've lost our ability to think big and build big.

Slowing down is deadly. When you have slow to stagnate growth, mold forms on the foundations of empire and poisonous philosophies and ideas and people move in to fill the void. A faster, growing economy makes people happier and gives them more choice. When your company fails you just go work for the winner or get a better job. But when you've got an economy that's just treading water every job loss hurts and leads to rising unrest and resentment and fuels a new wave of bad ideas and bad policies that makes everything worse.

To fix it we've got to battle back against the dark ideas of degrowth and the precautionary principal and the AI doomsday fanatics, as well as the anti-science, anti-tech mentality that's taken root deep in the psyche of the country and killed our ability to dream of a better tomorrow.

When you throw a wrench in the wheels of progress, bad things happen. Enemies of innovation always think they're on the side of the light by slowing things down. But they never think about the side effects, about all the good things that we miss out on when we gum up the works. What diseases will we cure with stem cell breakthroughs decades too late because we wasted eight years in the second Bush administration restricting the research based on worries about aborted fetuses?

The National Library of Medicine reports on the drastic slowdown in progress that came out of the Bush era decel policies:

"A slew of negative ramifications followed for ES cell researchers. Now facing restrictions on the type of research conducted using federal funds, some scientists were forced to create a dichotomous research environment based on federal vs. private funding of staff, equipment, and lab space. Collaboration and the sharing of knowledge between scientists also was hindered, and American researchers who previously spearheaded ES cell initiatives were no longer able to offer much of a contribution, stifling relationships with their international counterparts.

Further aggravating the situation was the fact that the 21 existing lines were not genetically or ethnically diverse, meaning specific disease processes (such as Parkinson’s) could no longer be studied in ES cells. Similarly, any information gleaned from the existing lines was limited to certain ethnicities, leaving uncertainty with regard to cellular processes in minority groups. In terms of therapeutic application, all 21 lines were of decidedly poor utility as they were cultured under inferior conditions by today’s standards."

This was and is sheer stupidity at every level.

What's the cost in people dying now, who could have been saved by advanced progress on stem cells earlier? What's the benefit of slowing down, except cheap political wins and appeasing people who've never solved a real problem in their life? I'll trade the hurt feelings of a bunch of religious political activists for saving the lives of children with genetic diseases any day of the week.

But there's equal share of blame to go around on both the left and the right.

You can blame the climate crisis directly on the environmentalists of the 1960s. They attacked nuclear energy and left us with ever rising emissions because of much dirtier alternatives.

Luddites always think they're saving the world. If they can just slow down, stop or break the engine of progress they can take us back to a golden age where life was simpler and everyone knew each other and everyone lived off the land. They forget that those were times of backbreaking labor and lack of choice, where half of all children died before reaching adulthood and a tiny cut could mean the end of you.

You've got to have respect and reverence for the past. Life is not all about growth and economic wonders and new technology. Family, friends, good food, laughter and love are all the things that make life wonderful and worth living. But technology and progress mean a longer life for the ones that we love. It means we get to see our children grow up. It means you can get on a plane and see Japan, instead of just reading about it in a book. It means that your country is safer and can defend itself and adapt to crises on the fly because every dollar is not stuck just paying for the basics.

It also doesn't mean that we never need checks and balances on technology. Seatbelt laws were good and saved a lot of lives. Before seatbelts were mandatory, you had to buy them on your own dime and most people just didn't bother. Laser focused and highly targeted legislation can work wonders.

Unfortunately, we've progressed towards bigger and more bloated bills. The Glass-Steagall Act was 37 pages. The Dodd-Frank bill that eventually replaced it was 847. It's hard to get anything big done in a world strangled by red tape that requires an army of $3000 dollar an hour lawyers to interpret.

To build a better future, we've got to find a better balance. We don't need sci-fi that's just fawning over technology and never shows the drawbacks but we do need something much, much more than a black mirror on the world. We need clear eyes and a clear window. We need something to believe in.

If you're alive today in America, you're one of the luckiest 1% of 1% to ever live. You've lived longer, healthier, with more choices and more freedom than at almost any other time in the past. But you've also lived in a time of increasing uncertainty and instability, where the old foundations are crumbling the answers to fix it are increasingly insane and wrong.

There's really only one answer and it frames everything you need to know when you think about the environment, laws, policies, jobs, technology and the future.

Build.

We need an American DeepSeek.

We need a sovereign, open source AGI powerhouse and we need it now.

That's because American frontier labs are losing the race.

Don't believe me?

For every one American closed or proprietary model, we're seeing dozens of industrial grade open models out of the Middle Kingdom.

The only American open source champion, Meta, put out the last Llama model, Llama 3, a year ago on April 2024. It's since been swept away by a tsunami of Chinese models. A year is a lifetime in AI. And even worse, its subpar multimodal capabilities don't hold a candle to the multimodal power of Chinese models.

Llama 4 will likely deliver amazing results but one American powerhouse can't keep up with the breakneck speed of releases needed to compete against 4 or 5 or 10 open source AI companies out of China dropping models every month.

Since Llama 3's initial release we've had DeepSeek V2 and V3 (and a new V3 iteration just this week), as well as the shockingly capable DeepSeek R1, along with Qwen 2 and 2.5, powerful text only models, and Qwen 2.5 VL (probably the most powerful multimodal model available right now outside of the closed alternatives, trained on everything from video, to GUI navigation, to agentic tasks), with updates like the 32B model just dropping, along with models from ByteDance and more, not to mention the absolute storm of video models out of China that have absolutely crushed Sora.

The new LLama is coming fast but the rumor mill was churning after the release of DeepSeek R1 that Meta engineers had scrambled a war room to tear apart its papers trying to figure out where they went wrong with Llama 4 and it's since been delayed while they catch up to R1's reasoning capabilities.

Think American chip sanctions will keep the Chinese AI juggernaut at bay? They won't. China is pouring billions and billions into creating cheaper, faster, better EUV light machines and they're testing them right now. Almost certainlty, the chip sanctions will be utterly worthless in just a few years time. All they'll do is guarantee the rest of the world is buying Chinese chips instead of American ones.

Something has got to change and it's got to change fast.

We need more American open source champions and we need them now.

Even worse, not only are American labs not releasing open source, sometimes they're not releasing at all.

Sora came out to great fanfare on X but then didn't come out for almost a year later, when it had already been outclassed by other video models at home and abroad. While American labs have been infested by EA driven "safety" pogroms and obsession with centralized control, Chinese companies just keep releasing at a furious pace and laughing as we shoot ourselves in the foot.

There are multiple reasons for American frontier lab stumbles but the biggest is capacity. GPUs are massively expensive and nobody can buy or rent enough to meet growing demand. OpenAI dropped GPT 4o image gen and a few days later Altman tweeted their GPUs were "melting." Less than a week after the release of Gemini 2.5 the head of Google AI Studio said they were still "constrained by rate limits" giving developers a measly 20 requests per minute. No single entity can seemingly keep up with the inference demands of mega-models.

Sometimes the reasons are commercial. o3 made massive waves by smashing the first iteration of ARC-AGI but where is it? Supposedly it powers DeepResearch but nobody else outside of OpenAI can get their hands on it, just smaller distilled versions like o3-mini. They'd rather it power their tools than let you get your hands on it to build a competitor to DeepResearch.

Other times it's just the size of these models, which points back to capacity. Newer models like 4.5 are huge, trained on 10X more compute, and yet they're prohibitively expensive and basically no better than their older brothers. The pre-training scaling laws are dead.

Sometimes it the culture of anti-tech and degrowth that’s poisoning American innovation and its historical optimism. The big American labs have been positively infected with people who want to stop, slow down or cripple AI with chains and guardrails or just stop it from happening at all.

(Source: Neil Christensen and GPT-4o)

Meanwhile, Chinese AI powerhouses like DeepSeek, ByteDance and Alibaba have ripped a page from that old American playbook of global software dominance:

Win with open. Win with soft power. Win with good enough.

Is 10 points of difference on a benchmark worth 100x the cost to you?

Nope.

Not for 80% of the use cases.

And while American companies struggle to meet demand, Chinese companies are releasing like crazy.

If you release an open model you get the benefits of distributed scale. People can run it on their own GPUs. Since nobody can get enough GPUs to power all that demand, better to just release it and let the power of distributed GPUs power it. Yeah you don't capture all those dollars yourself, but you capture mindshare. And some of those dollars come back to you. When someone builds an enterprise app on an open model, they're likely to come back to you for faster enterprise inference and Max versions of the model to power that app in production, so its a virtuous circle.

Open is the winning strategy now and it will be the winning strategy tomorrow.

Over the coming decade, open models will form the bedrock of smart economy and we'll see a wild array of closed and open apps built on top of them.

Remember, open doesn't mean no closed apps. It means we commoditize one layer of the stack and then move up the stack to build more interesting things. Twitter is closed but it couldn't exist without open, commodity building blocks like Docker, Kubernetes, Linux and Pytorch.

But how does that even work? How do you give away software and make any money at all?

While open source companies don't profit from every deployment, the playbook for open source has always been apps built with commodity bricks, enterprise support and services for software, which is why companies like Red Hat, Hashicorp and MongoDB became multibillion dollar businesses while "giving away" their core products.

When it comes to AI that model will play out again.

Open source companies will commoditize intelligence and we'll move up the smart software stack.

We'll see enterprise services like "long context" versions of the models, finetunings for specific use cases, inference scaling, memory, apps, APIs, SDKs, and consulting to drive revenue around those models. We'll also see the open core model take root, where some parts of something like a Mixture of Infinite Experts model (where fresh experts or adapters are added to meet a specific use case) are open sourced while skillpacks of advanced capabilities are monetized. We'll also see sleek and integrated interfaces that people pay for and we'll see platforms for tuning and training the models even more.

DeepSeek was the turning point for AI. It's taken off inside and outside of China, showing rapid adoption at companies like Perplexity, Amazon and enterprises everywhere.

Over the last two decades American-led open source swept across the globe, creating trillions in economic value, powering everything from the phone in your pocket, to the router in your house, to American mega-clouds like AWS, Azure and GCP, to 100% of the most powerful supercomputers on Earth.

On a pound for pound basis, nothing creates value like open source at the civilization level. As Clem Delangue of Huggingface wrote recently about the Harvard study on open source:

"$4.15B invested in open-source generates $8.8T of value for companies (aka $1 invested in open-source = $2,000 of value created) - Companies would need to spend 3.5 times more on software than they currently do if OSS did not exist."

Chinese companies have learned from it. American AI labs have forgotten it.

American labs have gone totally closed. They benefitted immensely from open research and open designs like the Transformer and Pytorch and Linux and Docker containers and Ray and yet they've gone in the exact opposite direction.

While American labs focused on closed source models while trying to make competition illegal by banning Chinese models from the American market and pushing for draconian chip export controls, Chinese labs keep winning with open.

But with all that said even open is not enough to win the race to AGI.

We need something different. Something more. We need to solve the biggest problem in AI.

And what's that?

How to make AI profitable.

Not only are our labs not profitable, they're bleeding money. Anthropic jumped into the Google anti-trust case because the government was looking to block the search engine giant's ability to invest in AI companies and Anthropic basically told the government they needed the money to keep running. Despite a $200 pro subscription for their elite models, Altman said OpenAI is losing money on it. That's no surprise when a single new cutting edge GPU costs 30-50K a pop.

Despite having billions in investment and scores of eager customers around the world, they aren't turning a profit.

This can't last.

We need a brand new business model in AI, one that nobody's figured out yet. And what kind of business model?

One that solves the hardware problem.

What's the hardware problem? Only the hardest problem in AI right now. It's not new algorithms or how to make a better, smarter, faster model.

It's how to deal with the fact that GPUs are freaking expensive and they're nothing but a massive cost sink to our AI labs. They're a catch 22. They're essential and yet they're the reason labs keep losing money.

Companies raise and spend billions on fleets of GPUs that depreciate faster than a brand new car. Software companies need them and yet they're nothing but a nasty drag on the bottom line of every single AI company on the planet.

So far, in this early era of AI, the only person getting rich is Jensen.

So how do you solve the Catch 22 of the AI era?

We take a page from another playbook.

Amazon.

The AWS of AI

To solve the problem of AI hardware, you've got to flip that hardware from a cost sink into a profit center.

To do that, tomorrow's frontier AI labs need to transform from software only companies into vertically integrated software and hardware companies.

One way to do that is to adopt the AWS model.

So what is the AWS model?

Get incredibly good at building datacenters and building a platform around those services and then rent out that platform to create the "operating system" of the Internet. That was the original idea behind the first cloud service.

AI will become the intelligence layer of the Internet. Think ambient intelligence: a billion thinking machines plugged into billions of smart apps and services, threaded through every aspect of our lives, from the supply chain, to manufacturing, to defense, to entertainment, marketing and more.

There's a persistent myth that AWS got started because Amazon had a lot of excess capacity hanging around. It's wrong. The real reason AWS got started was that Amazon was a low margin business and because of that they had to get really good at building data centers and running them profitably.

"As the team worked, Jassy recalled, they realized they had also become quite good at running infrastructure services like compute, storage and database (due to those previously articulated internal requirements). What’s more, they had become highly skilled at running reliable, scalable, cost-effective data centers out of need. As a low-margin business like Amazon, they had to be as lean and efficient as possible," writes Ron Miller in TechCruch, about How AWS Came to Be.

They also wanted to fix their tech debt. They'd grown massively and wanted to untangle the mess of distributed software they'd build as they grew into a clean, API driven, consumption architecture.

"At that point, the company took its first step toward building the AWS business by untangling that mess into a set of well-documented APIs. While it drove the smoother development of Merchant.com (a now defunct platform), it also serve the internal developer...too, and it set the stage for a much more organized and disciplined way of developing tools internally going forward.

“We expected all the teams internally from that point on to build in a decoupled, API-access fashion, and then all of the internal teams inside of Amazon expected to be able to consume their peer internal development team services in that way. So very quietly around 2000, we became a services company with really no fanfare,” said former AWS CEO (now Amazon CEO), Andrew Jassy.

For AI companies to get profitable they're going to have to do the same. They'll need to rent out GPU capacity at scale, while creating a slick, hyper-valuable software layer over the top of it all. AI isn't a low margin business, it's a negative margin business right now.

Think CoreWeave meets OpenAI, if OpenAI were truly open.

The strategy is to buy mega-clusters of GPUs, then reserve half of those clusters for training but rent out as much as 80% of the reserve when the lab isn't training or doing experiments, both as reserved instances and spot instances.

That makes a truly cutting edge frontier lab that can pay its own bills in short order and has a constantly monetizable supply line of advanced chips both for training and for a hardcore on-demand HPC business. It flips a GPU into an asset instead of a CapEx and OpEx monstrosity. More demand means these labs can buy more compute with debt financing. If you've got enough customers and more clamoring to get on board the banks will give you money to go buy more chips to deploy.

And it can't come fast enough because a brutal reality for America is looming fast.

The reality is we're not ahead. We're falling badly behind it and it won't be long before we're too far behind and Chinese models have become the standard for the world.

While American labs will still be arguing about how the government can save them from competition with chip controls and bans, China will have already developed their own EUV machines (it's happening much faster than American analysts think [they're still thinking 5-7 years and they’re going to be dead wrong] and they're going to get blindsided by how fast) and they'll be using their superior manufacturing at scale capacity to crank out a blistering array of high end chips and they'll flood the market with chips and models and intelligence that the world will lap up because its cheaper and good enough and eventually state of the art.

Every country we slighted by putting them on the tier II or tier III list for export controls will be avid buyers of exclusively Chinese chips.

(Source: X)

You see, the world doesn't care if the future of models is Chinese or American.

Only Americans care about that. The world will use whatever works and works well. If it's American, great. If it's Chinese, no problem. If it's Ethiopian, all good.

So if Americans want to win there's only one path forward:

Compete.

Either we build the better mouse trap or we watch as others build it from the sidelines.

There's still time.

But it's running out.

The Brutal Reality

The brutal reality is that America is losing in both AI and robotics. We're losing in the digital world and in the physical one.

It may seem like we're winning with AI but we're not even close.

Right now, Chinese companies are releasing a storm of highly capable models that are almost as good as their American closed counterparts at a fraction of the price. In some cases those models are even better than the closed version, most notably on agentic capabilities like navigating GUIs.

If there are no American alternatives or those American alternatives are a year old, American companies will build with Chinese models. It's already happening now. The secret weapon powering most American agentic companies right now are Chinese models like Qwen.

That's because Chinese models are highly tunable and controllable, unlike their closed alternatives. When DeepSeek R1 rocked the world on its release, a gaggle of people who are absolutely clueless about machine learning shouted to the high heavens that the Chinese models are censored. Just ask them about Tibet they screamed!

They were wrong.

Perplexity (and anyone who knows the 101 basics about ML) knows that you can remove that kind of bias in days with a little fine tuning, making them more tunable than their American counterparts. Perplexity dropped an uncensored R1 on HF in short order and it now powers the reasoning of their search engine.

(Source: Perplexity blog announcement on DeepSeek)

Try to finetune Claude or GPT to change its biases and watch how long it takes or how impossible it is because of their layers and layers of "safety" and censorship controls. Even if you manage to fine tune it, you'll probably run into their traditional software layers that try to protect against certain use cases but end up hitting you with false positives instead.

"Safety" is a tax on downstream systems.

In the early days of o1, the filter layers at open AI often falsely flagged my questions about my own personal machine learning experiments on reasoning, probably because these dumb safety layers saw my questions as an attempt to subvert their rules that hide their internal reasoning token traces despite me and my team doing nothing of the sort. That makes for a bad user experience. Luckily, they seemed to have learned from their mistakes and the behavior went away, but it doesn't mean I won't see false positives in other scenarios and it's a bitter reminder that if you're playing by someone else's rules they can rip anything away from you and shut you off over night.

The recent GPT 4o image release is a perfect example of what happens when a few companies control what you can and can't do with your models. The first few days people where happily releasing a flurry of Ghibli inspired images. But a few days later they were already censoring Ghibli generations out of fear of a lawsuit.

As more and more Chinese models come out, more and more companies around the world will build with them, because they're cheaper and more tunable. They're not just good enough either. They're often better. Those models will gain mindshare and economic might. A robust ecosystem of fine tunes and additional capabilities will spring up around these models and before American companies know what hit them it will be too late.

When it comes to robotics the news for America is even worse off than that. Robotics and AI are closely linked. If models are the brains, then robots are their bodies in the real world.

In a brutal wake up call to the west called America is Missing the New Labor Economy, SemiAnalysis shows that America and the EU are badly behind on roboticization and the race to general purpose robotics.

When it comes to the sheer number of robots deployed in Chinese factories, nobody else in the world is even close. Take a look at that chart below. Even an idiot can see who is winning that race.

(Source: SemiAnalysis)

The authors paint a dire picture for the west:

"This is a Call for Action for the United States of America and the West. We are in the early precipice of a nonlinear transformation in industrial society, but the bedrock the US is standing on is shaky. Automation and robotics is currently undergoing a revolution that will enable full-scale automation of all manufacturing and mission-critical industries. These intelligent robotics systems will be the first ever additional industrial piece that is not supplemental but fully additive - 24/7 labor with higher throughput than any human—, allowing for massive expansion in production capacities past adding another human unit of work.

"The only country that is positioned to capture this level of automation is currently China, and should China achieve it without the US following suit, the production expansion will be granted only to China, posing an existential threat to the US as it is outcompeted in all capacities.

"This is the manufacturing playing field that China has dominated for years now. The country has one of the most competitive economies in the world internally, where they will naturally achieve economies of scale and have shown themselves to be one of most skilled in high-volume manufacturing, at the same time their engineering quality has grown to be competitive in several critical industries at the highest level. This has already happened in batteries, solar, and is well underway in EVs. With these economies of scale, they are able to supply large developing markets, like Southeast Asia, Latin America, and others, allowing them to extend their advantage and influence.”

Robotics is about to explode with a force that dwarfs past industrial shifts. Picture factories churning out robots that build still more robots. Each iteration cuts costs, boosts quality, and powers a production flywheel that keeps spinning ad infinitum—leaving rivals more and more desperate and breathless as they sprint to catch up but find out its hopeless.

Because robotics is a true general-purpose technology, it will upend everything from textiles and electronics to consumer goods.

Meanwhile, the West is about to get caught flat footed as the unexpected punch smashes into our face. Here's a fully automated factory in China cranking out the components for a thousand missiles a day 24/7. Meanwhile American factories are still producing missiles like its 1945, by throwing people at it. A Pentagon war simulation said the US runs out of missiles "in days" in a war with China because it can't match China's production capacity. Europe’s industrial core is getting devoured by China’s relentless rise and its absolute inability to source enough power (while shutting down their nuclear reactors), and the US keeps fixating on other markets while banking on cheap offshore production.

(Source: Weibo, “dark” factory in China)

All the while, China’s manufacturing might gets stronger, its self-reliance grows faster, and robotics and intelligence are entering a bold new age.

This is not the recipe for the west to win and if we're not careful the next century is the century of the Middle Kingdom.

Even worse, we're asleep at the wheel. Americans think we're ahead with AI and robots and everything else. We imagine it's our birthright to be number one forever and forget that empires are born through a combination of constant innovation, new technology, creativity, acceleration, trade, business and brilliance.

That's how empires fail.

If we're not careful, we'll suffer the fate of every empire before us.

And that's because we've stopped seeing reality and we just see what we want to see. We're living in a daydream of our glory days in our heads. When the alarm rings on the new world, it's going to be a nasty surprise.

But all is not lost. The game is still being played and there's still time to turn it around.

To do it, we've got to wake up and do all of this and more:

• Greenlight bold new nuclear energy initiatives at breakneck speed

  • (China has 150 new reactors planned by 2035)

• Deliver massive investments in critical mineral mining wherever those deposits are in the world so we can make chips and everything else electronic

  • (Where critical minerals are)

• Invest heavily in a public-private hybrid for advanced chip making capabilities, following the original TSMC playbook where the government clears roadblocks, lowers taxes for the chip giants and greenlights new energy sources

  • (History of TSMC)

• Ramp up investments in next-gen robotics

• Build half a dozen open source intelligence powerhouses

This is not a job for any single company. Nobody can do it alone.

We don't need more closed AI companies. We need a dozen vertically integrated ones that can massively scale hardware and software to dizzying new heights.

We need automated factories by the hundreds.

We need robots, lots and lots of robots.

There's some hope for American robotics already, some bright lights in the darkness. We've seen a huge investment by venture capital into humanoid robot companies like Figure and humans are good design. They may just be the design that gets us to universal robotics. Robots so far have been highly specialized designs that do one thing and do it well. But those robots are brittle and they can't adapt. If the riveting sheet is off by a small fraction of an inch the robot riveter arm fails. Humanoid and human spin off designs hold the possibility of adaptability.

But when it comes to open source AI labs in America we've got Meta and nobody else. That's not enough to stand against the tide of Chinese models coming out with blistering speed.

Even worse, our companies can't seem to make a profit as they choke down 30 or 40K GPUs that eat profits like a great white shark. We need profitable open labs.

One company can't do all of that and nobody should even hope to try, because it's a fools errand.

But it has to start somewhere. And the best place to start is a lighthouse American open intelligence lab, a brilliant light on the hill that inspires other American companies to follow suit, one that's cracked the hardware problem and makes amazing software too.

Do that and we might just turn it around.

Let's make AI open again.

While we're at it, lets solve the profitability problem for American's labs.

And if we do that, we just might see the American dream reborn for another day.

If Linux were just starting out today, it would get crushed.

It would get killed by lawmakers, skeptics, doomers, alarmists and vast array of hysterical conspiracy theorists spinning terrifying fantasies of what happens if we let software be open and free.

They'd scream that open source was enabling our enemies and it's too dangerous to let into the hands of regular people and it's got to be tightly controlled or really bad things will happen! They'd spend billions and fire up idiotic law after idiotic law at the state and federal level to make Linux kernel developers responsible for "harms" aka whatever nebulous, ill-defined, anti-democratic, anti-choice thing they could imagine.

And the entire world would be so much poorer and worse for it.

Why? Because Linux and open source is the invisible backbone of the digital world.

It powers 100% of the top supercomputers in the world, the clouds of Amazon, Microsoft, and Google, US army laptops, many US Navy sonar systems, the router in your house, your smart phone, Ham radios, the Playstation 5 (powered by FreeBSD, another open source OS), and so much more.

A report by a team at Harvard Business School estimated the value of open-source software to be around $8.8 trillion annually to the global economy. Rebuilding Linux’s codebase and infrastructure would require $4.15 trillion (supply-side) to $8.8 trillion (demand-side) in investment.

But if Linux were starting today it would get murdered before it even crawled out of its crib. Trillions of dollars in economic value would go right up in smoke.

And that's because today, there's just too many hostile forces arrayed against openness.

If you tell today's politician that we'd get those trillions in economic value plus the strategic hard and soft power advantages of having the entire world run on our software but China could use it too, they'd say no, we we've got to stop it, not understanding that there is simply no way to get all that without having it open in the first place.

Open is the very essence of what drives all this value and there is no way to get it with partially open, or kind of open, or sort of open, or just outright closed.

How do we know? Because we tried semi-open in the early 90s with encryption and it simply didn't work.

US encryption export controls in the early 1990s crushed early e-commerce development by forcing companies to use weakened security measures. That's because short-sighted Cold War-era policies classified encryption as munitions, and required export licenses for cryptographic tools exceeding 40-bit key lengths. Browser makers like Netscape had to release a 128 bit encryption version for Americans and a weaker 40 bit encryption version for everyone else.

(Source: San Francisco Examiner)

Unfortunately, that just ended up compromising American security too. Why? Because the hoops you had to jump through to get the high encryption browser meant that many Americans didn't bother and just used the weaker version too, putting millions of people at risk with easily and laughably hackable "security." These rules weren't reversed until the late 1990s.

Until we had strong encryption worldwide e-commerce simply couldn't take off because who would trust their credit card to something insecure? Outside of browsers, it also cost America billions in foreign sales to our allies and friends in places like the EU and the UK. Banks and other security demanding industries overseas rightly demanded better encryption and so turned to local software makers who could offer them the security they needed to run their business right.

Today, global e-commerce is between $6.56-$7.4 trillion annually, demonstrating exponential growth despite these early obstacles.

How much more would it have grown and how much faster if we'd taken the chains off much earlier?

This is what short-sighted, alarmist, jingoistic policies do. They hold back the whole world and they blow back on the country making them too.

Luckily when Linux was getting started, it was a different world. It was a world where software was free to develop naturally and organically without much interference from a droning chorus of activists and antagonists because few people understood technology and there was no social media to amplify alarmist's voices.

Beyond that, the US had just won the cold war. It was the one true power on Earth with no realistic rivals. China was still a poor backwater, crawling its way out of the disastrous Mao era and ruinous communist social experiments like the Great Leap Forward and the Cultural Revolution. China's GDP in 1991 (the year Linux was released) was $383.37B. By contrast, the US had 6.158 trillion in GDP in 1991 and China had 18.80 trillion GDP in 2024 (with the US at 29 trillion in 2024).

The first Gulf War shocked militaries around the world and everyone feared the awesome might of the US military which rolled over the 4th largest army in the world like it was a video game in a few months.

(General “Stormin’” Norman Schwarzkopf, architect of the first Gulf War)

The software tech industry was small and the average person didn't have any opinions about it, much less any understanding of the web. Politicians like Clinton embraced tech as an engine of growth and prosperity (with notable exceptions like the v-chip). Lawmakers knew they didn't know much about technology and generally stayed out of it, letting it develop naturally and organically, with a few minor exceptions. No one would pass Section 230 today. Despite it having made the modern web what it is today it's hated on both sides of the isle in American power politics.

Today we have a much, much different world. It's a multi-polar world, with China ascendant and rivaling or surpassing American economic power in multiple areas. China's car industry went from non-existent to the number one car manufacturing engine in the world in just twenty years, producing 30 million vehicles a year, double what the US produces and at 20%-25% of the cost of the rest of the planet. DeepSeek shocked the world and showed many Americans that China wasn't behind on AI at all and may even lead it soon, if we're not careful.

(Source: BofA Global Research)

But the biggest difference today is that we have a whole range of people, from lawmakers, to social pundits, to the local cab driver with zero understanding of software, AI, and technology and who take incredibly aggressive stances on it.

Hundreds of bills are pressing forward at the state level, creating a tangle of liability, paperwork and rent seeking compliance officers who produce no economic value by driving up costs by 5-15% across the board. We're averaging 19 new bills per day and looking at over 1000 bills by the end of 2025. Make no mistake, the EU AI Act is Coming to America.

And that's because for over two years, this band of doom-sayers and alarmists have hijacked public discourse—spinning sci-fi doomsday fantasies fueled by personal insecurities, convinced that they'd unearthed humanity’s ultimate threat and they were the only ones who could see clearly and stop it!

ChatGPT was a catalytic moment for Doomers and alarmists of all stripes. They seized the moment as ChatGPT's exploded onto the world stage to shout that AI was about to outsmart us all and we needed to act before it was too late!

The alarm spread like wild fire and they fanned the flames of a full blown moral panic.

Unfortunately, too many people, especially lawmakers, listened and the fallout was a bunch of clumsy, shortsighted policies like the EU AI Act and boondoggle events like UK’s AI Safety Summit, a virtual circus of grifters and charlatans taking about utter nonsense like AI snuffing out all life by causing solar flares or sucking all the oxygen out of the world by gene editing plants. I wish I were making both of those up.

Europe suffered the worst because it listened to some of this early insanity. They pushed out Thierry Breton, the architect of the EU AI Act, too late. The damage was done. Now Europe faces the looming arrival of the law and the overwhelming burden of outdated, restrictive rules that stifle innovation without addressing any real world, practical risks like autonomous weapons or mass scale surveillance.

But who are these ill-informed and highly aggressive anti-open source/weights, anti-AI alarmists and how do we defend against them? Broadly, these folks fall into one of four categories:

• The monopolist

• The jingoist

• The doomer

• The idiot

One could argue that the last category is really just a variation of the first three but the idiot deserves his own category since idiots produce a unique kind of argument that must be defended against if we're going to defend open source in general.

Monopoly Money

Let's start with the monopolist.

This is the most straightforward antagonist of open-source and open weights AI. They have a simple and clear motivation: money.

Now don't get me wrong, there's nothing wrong with money. I like it and so do you and if you're say you don't you're lying. But the problems start when the monopolist doesn't want to play fair with competition. They'd rather ensure the competition doesn't exist or gets locked out of the marketplace via the law so they can't compete at all.

Monopolists will often pretend to like open source or pay lip service to it, but in general they despise it for one major reason:

Open source tends to drive down prices.

Monopolists would love to keep the prices high at all costs because they have a tremendous investment in R&D and a large install base.

Sam Altman who's taken a decidedly closed source approach with OpenAI wrote in a Reddit AMA, "I personally think we have been on the wrong side of history here and need to figure out a different open source strategy. Not everyone at OpenAI shares this view, and it's also not our current highest priority."

In other words, lip service. Open source is great but we're not really going to make any tangible shift towards it because it's not a current priority and it would likely gut their current pricing structure over time.

OpenAI's biggest rival, Anthropic, is more openly monopolistic. Half of CEO Dario Amodei's recent essay on DeepSeek and China was about prices and how we have to act fast to make sure China doesn't become a real AI superpower so we need deep sanctions and chip controls to keep them out of the market. But his main concern was that cheaper, open models would be good enough, less controlled and less opinionated than Anthropic's models, which would slice into their bottomline while dressing it up an national security.

The other big reason monopolists hate open source is because open source tends to make software a commodity. It makes it widespread and ubiquitous. It's permissionless and doesn't require anyone to go through centralized choke points to get it. It lets anyone dream up any way imaginable to use it and just do it without begging for the right to do it. Which again, means it comes down to price. It's much harder to charge top tier prices when everyone has a common set of building blocks that makes it easier to move "up the stack" because the pieces below have become commodity prices, aka cheap.

DeepSeek is already eating into OpenAI's bottom line. R1 was finetuned by Perplexity to strip out the model's unwillingness to answer questions about sensitive Chinese history, like Tiananmen Square, (due to compliance with local Chinese laws) and it almost certainly powers the $20 a month rival to OpenAI's $200 a month DeepResearch. DeepResearch is an amazing product and I'm a subscriber and lover of the platform but Perplexity's Deep Research clone, using a commodity intelligence layer, is likely good enough for many people's everyday needs, which means less people sign up for the more expensive subscription that is out of the range of most people's budget anyway.

Let's face it, the average person simply does not have $2,400 of spare cash a year to spend on AI apps.

Cheaper intelligence means more choice for consumers and more options. That's good for everyone except the monopolist.

But since the monopolist can't come out and say what they really want they have to frame their argument differently so they usually swing like Errol Flynn into the Moral High Ground Fallacy. This is a super simple attack and it works by appealing to vague moral principles or platitudes like "fairness" or "justice" or "compassion" and deflecting away the weakness of their argument by framing their opponents as unethical scum instead of addressing the substance of the argument.

When it comes to AI those two moral arguments are usually "safety" or "China."

Safety is a clever word. It's a universal platitude that has some real meaning but that mostly just allows anyone to project their own feelings and ideas into the word. It's like if someone says the word "God." People say the word and think everyone else is hearing the same meaning they have but really what other people are hearing is something completely different. It's colored by their own ideas and faith or lack thereof and so the word immediately carries a distorting charge that warps its meaning in conversation.

The same is true for a word like safety. Who doesn't want to be safe? If I offer you a choice between extreme danger and sitting at home comfortably safe on your couch watching TV, you're going to chose safe every day.

But what is safe in AI? Is it simply the inability for the system to take dangerous actions? Is it the refusal to answer certain kinds of questions? Who defines those questions? If I am a writer and I want to know the chemical formula for a poison to make a killer character in my story more realistic, how will the AI know the difference? Should it be making those kinds of assessments anyway? Who gets to define the rules? How do we know they're right? It quickly becomes a quagmire and very soon you start to realize that safety and censorship have a whole hell of a lot in common. They might even be the same animal.

Safety in AI usually comes down to giant fantastical fears like AI launching nukes or bioweapons, the kind of nonsense that comes from Hollywood films and has nothing to do with reality. Any idiot who's asked "do you want AI to control nukes or launch a deadly virus" is going to say no, absolutely not. Who in the world would want such a thing? The main problem is that it's a nonsense question in the first place. AI can't do these things now. It's just imaginary made-up speculation that may be possible at some point in the future if someone was stupid enough to link up an AI control system to a nuke but that has absolutely no relevance to safety discussions today.

So you scare people with a vague future threat and get them to push for closed, tightly controlled systems, which plays right into the ready and waiting hands of the monopolists.

"You don't want unsafe AI do you? Like what if AI killed you and everyone you loved? You wouldn't want that would you? Well we can keep you safe! All we have to do is ban openness, ban competition and we'll deploy special, trusted systems that only we, the trusted people, can control to make you safe."

Beware of monopolists bearing gifts.

The other vague trigger word for monopolists is "China".

China is the bugaboo of the left and right leaning defense hawks alike. It's today's Soviet Union, a big bad guy to pin everything on. Amodei's essay called for ramping up export controls on US AI chips from NVIDIA in order to create a "unipolar world" where the US and its allies hold all the cards in AI versus a multipolar world where China is an AI superpower too.

(Source: Bloomberg, Export Controls Map by Biden and Trump administrations)

AI card sanctions don't and won't work.

Chinese companies can easily set up shop in any country on Earth because today's businesses are multi-national by default. Set up in the US, Europe or Singapore and they can train models with NVIDIA chips and run them to their heart's content.

And Chinese companies are not sitting still. In a striking parallel to earlier encryption export controls, where American companies were locked out of competing in foreign markets, we've blocked our American champion, NVIDIA, from selling into the Chinese market with increasingly drastic export controls, and we've created a void that Chinese companies like Huawei are leaping in to fill. DeepSeek inference is already running on Huawei Ascend 910Cs and Huawei is already previewing 920C chips that aim to compete with Nvidia's upcoming Blackwell chips in terms of power and speed.

This is a losing game of tit-for-tat trade war, where we block them from getting chips, they make their own and then Beijing hits back and block us from key rare earths like Gallium that we need for everything from our special forces night vision goggles in our military to circuit boards.

It's a war of attrition with no winner, ever. We will have a multipolar world, whether people like it or not and the only thing that increasing the trade war will do is lead us to a monopoly of closed source American firms who leverage the China fear to get restrictions on competition.

Remember that ultimately the base goal of the monopolist is to reduce competition and control the market outright by having the government pass laws that favor their business and crush competitors. This creates a small group of a few companies who can do whatever they want, even if it's not in the best interests of small business or consumers.

American's business and technological prowess over the last few centuries came from small businesses that were able to operate and innovate as they saw fit with minor restrictions that made sense (with notable exceptions like AT&T in the 1950s which was a state sponsored monopoly). Lots of smaller companies compete and eventually some of them grow into massive companies. Kill that in favor of a few mega-companies and innovation slows down to a trickle as innovation gets tightly restricted to the limited imagination and gate keeping of one or two companies.

Already, AI has become a closed environment where the big American AI companies release no papers with any useful details and provide nothing back to the research community that they benefited so strongly from in the first place. Without kubernetes, Ray, containers, Linux, pytorch, the Transformers architecture, Claude and GPT simply would not exist. It was open research and open sharing that got them to where they are today. Without those breakthroughs they couldn't have built their models, so when Amodei says they didn't get any help building their model, it's nonsense because their model is built on the back of countless open innovations.

Without open source AI many American companies would simply die off too. Take my small AI company, Kentauros, for example. Without open source models we'd be dead.

That's because we're building a product called AgentTutor that helps people collect the data they need to teach AI models how to do complex tasks on computers. We do automatic labeling of the data with foundation models. The terms of service of the large American AI companies would prevent us from doing this because they forbid using any outputs from their models for training other models.

Why? Again, because they don't want the competition. It's really as simple as that.

By having a big, powerful foundation model that can generate synthetic data to uplift newer, smarter models, they have the tremendous ability to hillclimb faster than everyone else while preventing you from doing the same and that prevents even good use cases like mine. I'm a part of little tech, not big tech. I'm not competing with OpenAI to build GPT. I couldn't even if I wanted to. Our models are designed to do one kind of agentic tasks really well.

Other use cases get smashed by monopolistic licensing too. What if I was building a medical model to detect cancer and using GPT to do annotations of those X-Ray and MRI scans? I couldn't do it. I'd be locked out by the terms of service.

Open source changes this. It lets anyone innovate and dream up new ideas that can change the world.

At the Paris AI Action Summit, even former Google CEO, Eric Schmidt, has pivoted and called for the US to focus on open source or "China will ultimately become the open-source leader and the rest of the world will become closed-source."

The ugly truth is: this already the truth. China releases open models and all the US models are closed, except for Meta’s models.

Every American University researcher and small AI business is dependent on Chinese open source models like DeepSeek R1, Qwen and Qwen VL and many more. That's because, other than Meta, American companies have abandoned open source. This is a looming disaster and it's one that some people not only don't want to avoid, they want to make it much, much worse.

And that takes us to the Jingoist.

Love It or Leave It

The jingoist has a very simple world view. Anything their country does is great. The jingoist is blind to any and all criticism about their own country. Their country is perfect and flawless always. It can do no wrong. Everything it does is always good and always the right thing, as long as it's "for the country."

This kind of thinking is behind the recent bill introduced by Republican Senator Josh Hawley, who's bill would criminalize use of DeepSeek and other Chinese models for Americans with jail time and fines of millions of dollars. This may be the single most idiotic bill introduced in a sea of AI related bills at the state and federal level over the last year and half, outpacing even the ill-fated California SB-1047 bill that was vetoed by the governor at the zero hour, despite pressing forward for months against massive opposition.

Unfortunately, he's not alone. The incoming Commerce Secretary, Howard Lutnick, said of DeepSeek: “Open platforms, Meta’s open platform, let DeepSeek rely on it.”

Leaving aside the fact that this is a completely wrong claim, because DeepSeek didn't rely on anything that Meta released at all, in any way, it's incredibly worrying that the incoming Commerce Secretary is this misinformed and doesn't seem to have even the most basic understanding of how open source works, why it works and why it was worth trillions to the world economy and millions of American jobs from coders to support.

Hawley's misguided and idiotic bill is music to the ears of the monopolist because they won't have to compete with open source models from China and they can jack up prices and muscle smaller American startups out of the picture, who rely on open source models because they didn't raise 10B to train a foundation model from scratch.

University researchers are particularly left behind without open source. Even prestigious universities like Stanford struggle to get the mega-computing resources they need for students learning the trade. The AI focused university that boasts industry AI Godmother, Fei Fei Li, as part of the its faculty, is just now spinning up a cluster of 248 H100s, a far cry from the 10s of thousands, 100s of thousands or even millions that US AI companies are working with now and in the coming years. Musk's xAI boasts 200,000 H100 and H200 chips.

In the US we only have one open-source champion, Meta. The bill would also put tremendous pressure on them not to release their models either, because China would get access to them too, which takes us back to our earlier realization that if Linux were born today it would have gotten strangled in the crib and never made it to becoming the bedrock of world software that it is today.

We need bills that do the exact opposite of this bill.

We need to let Americas make their own free choices about what models they use.

We also need to encourage more American companies to release open source and open weights models, not try to shut down the one that is releasing open weights models.

In other words, you counter influence with influence, not with restrictions.

Essentially, out of pure fear of China, the jingoist would kill off tens of thousands of smaller American startups, crush small business tech jobs and leave them with no alternative because they've criminalized open source.

(Breadlines in the Depression)

To counter the jingoist there is really only one argument that works: It must be framed as good for the country.

Luckily, open source is good for America. The jingoist, who looks at the world through a black and white filter of "my country good, other countries bad" can only be countered with the cold hard facts of what open means for their country. And here they are:

Open source creates trillions of dollars in value, powers American military and critical infrastructure and serves as the backbones of countless American companies big and small. It's used by as many as 99% of American fortune 500 companies. If AI is allowed to develop properly and it doesn't get killed off by ludicrous legislation then it will serve as the backbone of the next American tech and business revolution.

Let them know that if open source is such a security risk, why does "the US Army [have] the single largest installed base for RedHat Linux" and why do many systems in the US Navy nuclear submarine fleet run on Linux, "including [many of] their sonar systems"? Why is it allowed to power our stock markets (NYSE, NASDAQ) and our clouds and our seven top supercomputers that run our most top secret workloads and the gov clouds that power all our of government agencies most important services like tax collection from the IRS which runs many services on AWS Gov Cloud?

It's likely that Senator Hawley doesn't know the sheer number of American companies and jobs that would get decimated by attacking open source and open weights AI and the last thing he or any other American congressperson wants to do is kill jobs. That's really the key for every American politician or any politician anywhere. Don't kill jobs. You want policies that create jobs and open source creates jobs.

So that's the way to fight the jingoist. It's maybe the simplest argument of all. Make it good for America or whatever country your local anti-open source politician hails from and you've got the only frame that's necessary to change their mind.

The thing is most people just do not understand and are not educated about what open source means to the modern world. They do not understand how critical it is to everything from our power grid to our national security systems and to our economies.

The main reason they don't understand it because because open source is invisible.

It runs in the background. It quietly does its job without anyone realizing that it's there. It just works. It's often not the interface to software, it's the engine of software, so it's under the hood but not often the hood itself. It runs our severs and routers and websites and machine learning systems. It's hidden just beneath the surface. It's an economic and engineering marvel of the modern world.

Open source is the most critical software infrastructure on the planet, bar none. That's why DARPA set out to understand "the most important software on Earth" in 2022. They wanted to understand it to make it even more secure and to understand why it works at all. “People are realizing now: wait a minute, literally everything we do is underpinned by Linux,” says Dave Aitel, a cybersecurity researcher and former NSA computer security scientist. “This is a core technology to our society."

It's not just a nice to have. It's absolutely critical to the functioning of the modern life at every level and to American security and the security of its allies. It's everywhere, an invisible bedrock beneath our entire digital ecosystem, underpinning all the applications we take for granted every day.

Show them that and you open their eyes to the light of the real world.

And so we come to the doomers, people who struggle to see the real world in any meaningful way.

Doomsday

Over the last few years battling the doomers, I've learned one thing:

You can't counter them directly.

You can't argue with them or convince them that AI will not kill everyone and take over the world like a megalomaniacal Marvel villain. Like Churchill said "a fanatic is one who can't change his mind and won't change the subject."

(Source: Marvel Studios, Ultron, robot villains are tired tropes in Hollywood)

Instead you've got to counter their influence, especially on legislators.

So far, we're mostly winning. The doomers were notably absent from the Paris AI Action Summit, when just two years ago it was called the AI Safety Summit. We fought back against doomer inspired bills like SB1047 in California and got it vetoed. Vice President Vance opened his keynote at the Summit by signaling a new era and a major policy shift:

"I'm not here to talk about AI safety, which was the name of the conference just a few years ago, I'm here to talk about opportunity."

But the fight is not over. Despite their waning influence in the public and over legislator's minds, many doomer inspired bills are moving forward at the state level, as we saw earlier, and so the legacy of their all-out campaign to make AI fear front and center lives on like zombies in legislation that must still be mowed down.

I've written extensive counter-doomer arguments in my series of X posts called Destroying Terrible Anti-AI Arguments, Day One, Two, Three, Four, Five, Six. Don't waste them on doomers. Use them to help everyone else understand that they're a small, fringe group with a big megaphone.

In those posts I argued against, six major doomer fallacies:

1. "Would you open source a nuke?"

2. "What if AI released a deadly virus!"

3. "XYZ expert says AI is dangerous"

4. "My p(doom) ratio is X%"

5. "We need control frontier models like nuclear weapons with compute thresholds, model registration and inspectors."

6. "ASI will kill us all"

Here's the short version of one of the most common ones "would you open source a nuke"? — this is a classic case of begging the question. The question assumes AI is monstrous without bothering to prove it. It falsely compares a multi-use tool (AI) to a single-purpose weapon (a nuclear bomb).

AI is not a nuclear weapon. LLMs are not nuclear weapons. They're are not even close. A nuclear weapon is designed to do one thing and one thing only, kill as many people as possible. AI has a vast range of possibilities, from detecting cancer, to writing emails, to creating art, to doing tasks, powering robots that will clean the dishes and mop the floors, proving math theorems and more.

Very few technologies are inherently destructive like a nuclear weapon and so when you hear someone making that analogy you know they're making it from a lack of understanding or bad faith. They know it's not a real argument and they make it anyway because they have an ulterior motive which is usually to link AI to something horrific in your mind so they can justify restricting it and locking it down and censoring it.

A kitchen knife can cut vegetables or stab a loved one. 99.999% of people on the planet use it to cut vegetables and we don't hear cries to take kitchen knives off the market because one delusional nutcase stabs someone with it. We let the 99% cut vegetables and punish the person who stabbed someone.

AI is like that—a general-purpose tool that mirrors its user's intentions.

(Source: Swiss Army Knife from Victorinox)

Think of AI like Linux. As we saw earlier Linux powers supercomputers, smartphones, and routers, but it can also be used to write malware and launch DDOS attacks. Early critics, like Steve Balmer at Microsoft, tried to kill it by linking it to diseases like "cancer" or hated political ideologies like "communism." But we embraced Linux for its immense, widespread benefits anyways.

We don’t ban kitchen knives because a few use them to kill someone. We don't lock away every tool that might get misused. Instead, we mitigate risks through targeted, clear legislation that punishes the person who misused it, rather than criminalizing the tool itself.

I won't rehash all the anti-doomer arguments here. The key to understanding these counter-arguments is they're not to there convince the doomer. That's a waste of time. They're there to convince the lawmaker and the journalist and the everyday person who has never used ChatGPT that these terrifying black and white scenarios have nothing to do with actual reality.

Convince people seduced by doomer fear-mongering what the repercussions of those deluded belief systems will be. Let them know that if they let that fear take root, it will be a devastating loss for us all, because the revolutions in tech that made it so easy for people to get an education online, chat with our friends all around the globe, find information on any topic at the click of a button, stream movies, play games, find friendship and love are facing intense pressure by governments and activists the world over who want to clamp down on freedom and open access.

If the open source restrictionists win, we're entering an ugly world.

It's a world where your AI can't answer questions honestly because it's considered "harmful" (this kind of censorship always escalates because what's "harmful" is always defined by what people in power don't like), where information is gated instead of free, where open source models are killed off so university researchers can't work on medical segmentation and curing cancer (because budget conscious academics rely on open weights/open source models; they can fine tune them but can't afford to train their own) and where we have killer robots and drones but your personal AI is utterly hobbled and lobotomized.

Doomers are closet authoritarians. They believe in total and complete centralized control and that regular people like you and me can't be trusted to make our own decisions. If you're fighting their arguments you've got to let people know that it's openness and permissionless access that's behind the majority of the major technological, economic and societal revolutions of the last two hundreds years. Permissionless access. Permissionless innovation.

(Source: Soviet Propaganda poster of early pre-AI Doomer, Joseph Stalin)

What made the web so fantastic was an open, decentralized approach that let anyone setup a website and share it with the world and talk to anyone else without intermediaries. It let them share software and build on the software others gifted them to solve countless problems. It worked before beautifully. Why wouldn't it work again with AI?

Permissionless is the key driver of progress in the modern world and in history.

And it's the foundation of open source.

The basic idea behind open source is to give everyone the same building blocks, whether you're the government, a massive multinational corporation, a tiny individual academic researcher, a community sewing circle, or a small startup business.

When everyone has the same building blocks, its easier to build bigger and more powerful and more complex solutions on top of those blocks.

Take a technology like open source Wordpress. Wordpress made it super easy for anyone in the world to make and publish a beautiful website, even with limited design skills. It now powers over 43% of the web.

When I was younger it was hard to develop and publish a website. You had to do everything from scratch. But a wave of open technologies made it easier and easier as the years went by. We had the LAMP stack (Linux, Apache, MySql, PHP), an operating system, web server, database server and web scripting language, which gave many developers the common tools they needed to create more complex web apps. That, in turn, enabled the development of Wordpress on top of those technologies. Better and more powerful stacks followed, like the MERN stack, MongoDB, Express.js, React and Node, all open source technologies.

Each new layer of software makes it easier to build more advanced and more useful solutions on top of the last layer. It's the essence of progress and development. When open ecosystems are allowed to flourish and the enemies of open are finally beaten back once more, we'll see a flowering of new progress and innovations in a self-reinforcing virtuous loop just as we have in the past.

When you only have mud hut level technology you can only build one story tall mud huts. When you have a hammer and nails and standard board sizes you can build much taller and more robust structures. If you have steel and cement and rebar you can build skyscrapers stretching to the sky.

This same cycle of openness benefiting the world permeate the progress of the world again and again across many different aspects of life.

Take publishing:

When six publishers controlled publishing completely we had a flurry of the same kinds of books over and over again, all fitting a few basic models. Publishers kept 90% of the profits of writers. But the digital self-publishing revolution gave us a flood of amazing new writers like Hugh Howey, with those authors now keeping 70% or more of their profits. Howey almost certainly would have never made it through the gate guardians of the big publishers to become a phenomenon with Wool and that was the fate of many authors before him. Gated access strangles progress.

Or take business and economics:

When starting a company required the permission of the king, you had a small group of super powerful corporations, like the British East India company that was once so mighty it conquered India with a private army that was double the size of the British army. That's right, it wasn't Britain that conquered India it was a giant mega corporation. And we think we have powerful companies now! Amazon ain't got nothing on the British East Indian company.

When breakthrough laws allowed anyone to create a company, we had a revolution of small businesses and a massive uptick in wealth and economic progress and longevity that has driven down poverty over two centuries to once unimaginable numbers and created more wealth than anyone in the 1700s could possibly even dream of and more.

Progress has a massive ripple effect that benefits the whole world.

Don't believe it? Just look at the world only a few hundred years ago.

Historian Michail Moatsos estimates that in 1820, just 200 years ago, almost 80% of the world lived in extreme poverty. That means people couldn't afford even the tiniest place to live or food that didn't leave them horribly malnourished. It means living on less than $1.90 a day in 2011 prices and $2.15 in 2017 prices. Actually you don't even need to go back that far. In the 1950s, half the world still lived in extreme poverty.

Today that number is 10%.

Nearly half of all children used to die, in every country on Earth, into the late 1800s, when economies were slower and more stagnant and less open.

In most first world countries infant mortality is now single digits or a fraction of 1% and even in the developing world it's 4%.

To see just how stark that drop is, take a look at this chart which shows child mortality over two millennia. You’ll see that for most of human history it was a flat line of uninterrupted death and then it suddenly drops dramatically.

That's what progress does. It makes the world as a whole better. That's what openness, sharing and common building blocks do. They create possibilities for more progress and more ideas that build on the last ones. Once you have an understanding of microbes, you can build better defenses against them. Once you have microscopes, you can see those tiny menaces and understand if the drug you crafted to stop them is working. Each layer of knowledge and tools enables the next level of knowledge and tools. When anyone can build on the past and abstract those lessons to new domains we get new breakthroughs and new ideas and new businesses.

When you look back at the history of progress and openness and sharing and realize how crucial it is, it's easy to take it as a given that everyone can easily see just how important it is but they can't. The forces of closed and the enemies of innovation are a powerful chorus and they strike back again and again in history.

They believe in command and control. They believe its better to hoard knowledge and keep it secret. This is what doomers and other closet authoritarians believe and they get alone really well with the jingoists.

They imagine that if we just make something a little bit open, instead of all the way open, they we can prevent bad people from ever getting their hands on the best tools. But it never works that way.

That's because closed and gated systems have tremendous false positive and false negative rates. Gated systems are inexact, inefficient and they block many people you want to have access and don't stop many of the people you don't want to have access. Every time we try this in history we create the unwanted side effect of dramatically slowing down progress because information, ideas and software can't flow freely and get trapped between ugly, inefficient, governmental gates.

Crush doomers with facts and grounded, sane, clear policies and then watch their arguments vaporize like steam on a hot summer day.

And that takes us to the idiot.

The Idiot

The idiot is a special kind of enemy.

The problem is their arguments don't make sense on any level and we're wired to take an argument at face value, to assume there is at least some merit in what they're saying. But with the idiot there is nothing realistic or grounded in what they’re saying at all. There is no stable ground to stand on. It's just illogical nonsense from top to bottom.

If I'm shouting on the corner that the sky is made out of cheese and it's raining donuts, you don't need to try to logically pick apart the finer points of my argument because there are none.

Gary Marcus is one of the modern masters of idiotic arguments. He has a nearly perfect record of incorrect predictions and yet he's still regularly called upon by journalists and governments as an “expert” in AI, despite zero percent of the AI community acknowledging him as one.

Take this nonsensical Tweet he wrote the other day about the DeepSeek phenomenon:

"Congress needs to bring in Zuckerberg and LeCun to discuss how their unilateral open-sourcing decision rapidly undermined the US advantage in Generative AI."

This is sheer idiocy because DeepSeek model wasn't based on any Meta models or any Meta ideas. In fact, the Chinese model outpaced Llama, Meta's flagship model, in just about every way. Not only did the DeepSeek team not need anything Meta released, the Chinese team created a number of their own breakthroughs independently, including a novel, state of the art reinforcement learning algorithm, GRPO, that's on par with PPO (the current SOTA) but that uses a lot less memory and bandwidth and they trained a reasoning model, R1, that's on par with closed source OpenAI wonder model o1.

If you're looking for logic in what Marcus wrote, don't bother. But if you really want to follow the basic flow of his argument it's this: "Meta released open source; open source is bad; Chinese researchers must have copied what Meta did because Meta released open source; China only copies things; since I think open source is bad/dangerous, congress needs to call Meta to a hearing so that Meta can be stopped from releasing more open source because they're giving away our secrets."

In other words, neither the premise, the intermediatete reasoning, the conclusion, or the solution to the problem has any logic or merit whatsoever because they are all 100% wrong and based on nothing realistic in any way.

So how do you defend against an argument that has no basis in reality, doesn't reflect any known fact?

Unfortunately, the idiot is often the hardest person to argue with and their stupidity has a tendency to spread. Even a marginally intelligent person hears a stupid argument and says "well that's just stupid, no one will believe that" only to discover that the belief spreads like wildfire with other idiots. It seems self-obvious that what the person is saying is stupid and yet it's not obvious to many people and you often find idiots with massive social media followings. It's almost counter intuitive, the more stupid stuff they say, the more followers they get.

The challenge of countering the stupid is put perfectly by Italian economist Carlo M. Cipolla in The Basic Laws of Human Stupidity:

"[Smart and wise] people find it difficult to imagine and understand unreasonable behavior.

"An intelligent person may understand the logic of a bandit. The bandit’s actions follow a pattern of rationality: nasty rationality, if you like, but still rationality.

"The bandit wants a plus on his account. Since he is not intelligent enough to devise ways of obtaining the plus as well as providing you with a plus, he will produce his plus by causing a minus to appear on your account. All this is bad, but it is rational and if you are rational you can predict it. You can foresee a bandit’s actions, his nasty maneuvers, and ugly aspirations, and often can build up your defenses.

"With a stupid person all this is absolutely impossible, as explained by the Third Basic Law. A stupid creature will harass you for no reason, for no advantage, without any plan or scheme and at the most improbable times and places...

"...one cannot organize a rational defense, because the attack itself lacks any rational structure. The fact that the activity and movements of a stupid creature are absolutely erratic and irrational not only makes defense problematic but it also makes any counterattack extremely difficult—like trying to shoot at an object that is capable of the most improbable and unimaginable movements."

• Cipolla, Carlo M.. The Basic Laws of Human Stupidity (pp. 53-54). (Function). Kindle Edition.

In other words, the very definition of stupidity is someone who causes you a loss and potentially even themselves a loss for absolutely no reason at all.

So how do we fight that?

The best and only argument against the idiot's attacks on open source AI or anything else, is to simply dismiss them as absurd and let people know that just because the person might have credentials or a long resume their argument is pure and utter nonsensical crap that doesn't even deserve a response because it's like arguing about whether the moon is made out of cheese.

Simply say there is nothing to see here and move on. Pretend they don't exist. Don't give them any attention at all and counter their influence elsewhere.

Unfortunately, this strategy doesn't always work. And the reason is that sometimes the idiotic argument comes from a high status individual. That's someone who's otherwise credentialed or smart or revered for their accomplishments in other way.

Yoshua Bengio is masterful at making stupid arguments about the future of AI and he is also one of the godfathers of machine learning and his contributions to the field of machine learning can't be underestimated. Sadly, he recently fell under the spell of the doomers and now spends much of his time campaigning about how we might "lose control" of AIs or they might "take over."

With a high status person you can't just dismiss them as outright idiots like you can with charlatans who've contributed nothing to field. They must be taken more seriously. But there's another rule of the Basic Laws of Human Stupidity comes into play:

"Stupid is independent of any other characteristic in a person."

In other words, someone can be a Noble prize winner or a fry cook and still deliver an utterly idiotic argument.

Take this recent tweet from Bengio:

Here he argues that "Science shows that AI poses major risks." Science does nothing of the sort. Maybe science fiction does but that's not anything to base policy on. Stories are about one thing only, conflict. They don't reflect real life, only a single aspect of life and that makes them a terrible lens to see the world. You’re guaranteed to see life through a darker lens if you view it only through stories.

Science, on the other hand, is a way of getting closer to empirical truth by having multiple people across the world independently verify an observation. There is no science that accurately predicts the future of how society and technology develops.

Science might verify the chemical structure of a newly discovered material or verify the efficacy of a drug to combat a disease. But there is no science that shows that AI will rise up and take over the world or kill us all or anything of the like. There are rampant and wild speculations but none of this is based on any firm evidence of any kind. So Bengio is either being deliberately disingenuous here in order to lend the gravitas of science to his belief system about evil AI going rogue or he's just making an idiotic argument.

Here's another one from a snippet of one of his speeches:

He says that once OpenAI has a truly powerful model they won't release it and instead they'll use it to "basically wipe out other companies and other economies that don't have these [models.]”

This is another incredibly terrible argument. It starts off with a reasonable premise, that an AI company might be economically tempted to keep their most powerful models to themselves and not sell direct access to them. That makes sense. It's a real possibility. This was the case with Google for many years, where AI powered their products like search but users never got direct access to it.

But after that the argument leaps into the absurd and suddenly OpenAI is ravaging whole companies and economies with no intermediate steps. How did we go from 1 to a million in a single leap?

In other words, in his scenario nobody else on Earth has created a powerful AI, only OpenAI in a vacuum. Also no other companies can defend themselves with their own adaptations, there are no legal or societal pushbacks or mitigations and entire economies go under. This is absurd. There are always counter balancing forces to life. As we saw with DeepSeek R1, even when OpenAI tries to conceal their most powerful training breakthroughs, other teams can replicate it in a few months through reverse engineering and logical deduction.

In other words, nothing happens in a vacuum. There will be pushback and friction and other companies and people with breakthroughs in AI at the same time so this argument by Bengio is another swerve into the absurd.

Unfortunately, the problem with a high status person like Bengio making this arguments is people are prone to thinking that expertise in one area applies equally to other independent areas.

They do not.

Just because someone may have forgotten more about machine learning than most people will ever know does not mean they're the least bit qualified to predict how technology effects complex systems like human society over time.

Whenever you hear something like “so and so is expert in AI and they say AI will take over the world,’ the first question to ask yourself here is "expert in what? Just because you are an expert in the inner workings of AI, does not mean you are an expert in psychology, business, economics, history, or how technology develops and changes over time. That's generally the work of sociologists, futurists and historians (who look for patterns in the past and sometimes try to apply them to the future). And even then it's a very inexact discipline.

So the answer to defending open source against the high status individual's idiotic argument is to acknowledge that person's status and contributions to the world but to crush the argument itself and remind people that just because you know something about one thing doesn't mean you get automatic credit in another area.

And if the person is just a general charlatan and grifter, with no real status whatsoever, just ignore them and move on to counter their garbage with other people so their idiocy doesn't spread.

You Are the Revolution

Microsoft once tried to kill open source. They launched a massive anti-Linux marketing barrage and tried to link it to disease and communism. They failed and now open source is the most successful software in history, the foundation of 90% of the planet's software, found in 95% of all enterprises on the planet.

Imagine if they'd been successful in their early attacks? They would have smashed their own future revenue through short-sightedness and total lack of vision.

Yet here we are again, with a witches cauldron of people who want to smash open source and open weights AI out of ignorance, blind animal panic and fear:

• Ultranationlists who think that if China gets to use something open too then it must be stopped at all costs, including smashing our own businesses and progress

• Aging AI safety fetishists like Max Tegmark, who worry AI will get "out of control" who want to crush and control AI development at every turn because they fear the end of the world like some bad sci-fi novel

• American AI lab heads like Dario Amodei of Anthropic who would rather strangle the competition and create monopolies than compete fairly on an even playing field

They must be stopped. It's really as simple as that.

If we let monopolists and the delusional drive policy in the world, nothing good ever comes from it.

If you want a world where small AI businesses are locked out of competition with the big AI labs, where Chinese models are banned in America, the land of freedom and opportunity, where your models are censored and controlled and what you can and can't do with them is strictly controlled, then embrace the jingoists, doomers and monopolists.

But if you want the next multi-trillion dollar revolution in software to benefit the whole world, you've got to fight them at every step until they're stopped for good and the technology can develop naturally and organically.

Time has a way of handling these problems. Given enough time, earlier idiotic laws will get rolled back and fanciful arguments about the end of the world will fall away as people experience AI in the real world. R1 and o1 were released months ago and the sky has not fallen. When people's experience don't match up with the doomsday predictions they'll adapt easily.

(Source: Midjourney, No Chicken Little the Sky is NOT falling)

AI will just be another thing to tomorrow's children, like a tree or a rock or a smart phone. It will just be there and they'll use it and understand it. If you tell them that some people once thought it would kill us all they'll just look at you like you’re nuts and wonder what the hell you’re talking about.

Maybe you think that lots of people have tried ChatGPT or another advanced model but the reality is that most people in the world just haven't. As more and more regular people get their hands on AI and see its benefits and realize the sky isn't falling, and that no machines have risen up to take over and the world hasn't ended, public opinion will naturally shift over time away from crazy fear and insanity.

We also need more companies to get back to embracing open source.

Fantastic models like Allen Institutes Molmo and Olmo. Amazing Chinese models like Qwen and Deepseek. American models like Llama.

We need a lot more American companies to embrace open source, not fight it because they want to lock up control of their models for a decade. And we need new countries getting into the picture. German models. Indian models. South African models. Saudi Arabian models. Italian models. Models from everywhere. We need a common intelligence infrastructure that we can benefit from the same way we benefit from open source software today.

AI will deliver an intelligence layer for the world.

But we might never get there if the jingoists and the stupid and the monopolists have their way with us.

Today's critics of open source AI trot out the same tired old tropes, like "security" or "China = bad" and they tell us only a small group of companies can protect us from our enemies so we've got to close everything down again and lock it all up behind closed doors.

Don't believe them.

Remember that the ultimate definition of stupid is people who do damage to others without any benefit to themselves and maybe even a net loss to themselves.

• The jingoist would kill Linux so it never runs on Chinese supercomputers, ensuring that it never powers their own super computers as well, aka cutting off their own nose to spite their face.

• The doomer would kill AI because they’re fear based creatures who read too many sci-fi books and because they want to feed their own egos and feel like they’re the noble hero saving the world from the apocalypse.

• The monopolist just wants to make sure they get all your money and don't have to worry about anyone else competing with them on a fair and open playing field.

• The idiot just wants to cause damage to themselves and everyone else for no gain at all.

If the only people working on AI have war chests of 10B or 100B you're going to have a very sad AI ecosystem, composed of a few powerful players with very limited and narrow thinking. That's not a world that we want. That's not America. America is about robust competition and the fact that any person regardless of where they came from can make a go at it and make it big.

We want competition and we want more minds working on intelligence. We want an intelligence revolution. We want ambient intelligence everywhere, embedded into every layer of the world economy, making it faster, better and more efficient. We want AI that rapidly creates new drugs that cure diseases and solves centuries old problems.

And there's only one way to get there.

Open.

What if you woke up tomorrow and there were no more NVIDIA chips?

Not only that but you can't buy a PlayStation 5 even at 500X mark up on the black market, or an iPhone, or a shiny new MacBook. You can’t get a new car or truck either because manufacturing plants from Detroit to Dresden are ghosts towns, totally shut down, because they can't get any of the 1000-3000 chips they need to make their vehicles go.

Like Thanos snapping his fingers in the Avengers, it's all just gone.

(Source: Marvel Studios Avengers Infinity War)

This is exactly what could happen if China invades Taiwan tomorrow.

To understand why you just need to understand the answer to one simple question: How many chips do NVIDIA, Apple, Meta, Amazon, Google, AMD, Broadcom, Marvell, Mediatek and Qualcomm make themselves?

Answer: Zero. Nada. Not even one.

They design chips and they get made somewhere else. That "somewhere else" is one of two places on Earth: Taiwan or South Korea. In those places you have two companies that make those chips: Taiwan Semiconducting Manufacturing (TSMC) in, you guessed it, Taiwan, and Samsung in South Korea.

But it's not really an equal share. By most estimates TSMC makes about 90% of most advanced chips in the world.(1) They make between 40-60% of the less advanced chips too. These aren't the sexy core processors in your Mac M3 or iPhone or NVIDIA H200s, but the secondary ones that make your WiFI go and your washing machine work. In your smartphone, you have between 40-50 chips, in your car or truck 1000-3000 chips, in jet planes (1000 chips), and the list goes on and on.

To put that in perspective, 90% of advanced chips is on par with the American Standard Oil monopoly during the late 1800s into the very early 1900s, where J.D. Rockefeller and Standard Oil controlled approximately 91% of oil production and 85% of final sales in the United States.

The old world may run on oil. But the modern world runs on semiconductors.

Everything from your washing machine, to the phone in your pocket, to your oven and your shiny new SUV runs on sophisticated little semiconductors.

Because we make barely any of those chips in the US, an invasion would shut off their supply in an instant and put them totally at the whim of mainland China, a mainland we've increasingly tried to hammer with sanctions, bans and tariffs. As we put more and more pressure on China and try to strangle its AI ambitions with increasingly arduous export controls, in a back and forth trade war where China retaliates by banning rare earths and precious metal sales to the US, which we need for everything from our military night vision Goggles, to motherboards, to satellites, there is no end in sight to the short sightedness.

Of course, you've probably heard that TSMC and other companies are starting to build factories in the United States. That's right. It's just not enough. Not even close.

TSMC are building three fabs in Arizona.

By contrast, in Taiwan, TSMC has eight 12 inch wafer fabs in, six 8 inch fabs, and one 6 inch fab, with five back end fabs. Those fabs alone can produce 12 to 13 million 12 inch wafers a year (1) and many more in smaller sizes. "A 300-mm wafer can fit roughly 72 GB202 candidates (which powers the RTX 5090), assuming that one die measures roughly 31.5 mm × 24.2 mm." It might fit 100s of smaller chip candidates.

TSMC is currently building a whopping one fab in Dresden, Germany and those are for practically ancient 28nm, 22nm, 16nm and 12nm nodes, not the most cutting edge chips which are 4nm and 3nm and soon 2nm and smaller.

Intel is building two fabs capable of 7nm. That's in jeopardy now since they've been bleeding money badly and their CEO, who spearheaded the return to owning fabs, just got ousted after a decade in command of the ship. These two fabs alone cost 20 billion dollars together.

Samsung is building one fab on US soil, in Texas, capable of 4nm and 5nm production.

That's it for advanced chip manufacturing in the United States and the rest of the world.

We have a grand total of one working TSMC fab, with a second one on the way and another at some point five years from now, two upcoming Intel fabs capable of last gen 7nm tech, and one Samsung factory on route. So 4 cutting edge fabs and 2 semi-cutting edge fabs by 2030 outside of Asia.

The rest of the dozens of fabs being built in the US are for RAM and other memory chips. Good and useful but those are not advanced chip fabs.

So what do we have in Asia?

Samsung currently operates four advanced semiconductor fabs in South Korea focused on cutting-edge nodes (4nm, 3nm, and 2nm), with additional facilities under construction. They are behind in total capacity versus TSMC for now. TSMC retains a significant lead in total wafer production capacity, but Samsung is aggressively expanding to close the gap with a combination of national and private funding. Samsung is gunning to be the top manufacturer in the world and they're working on the Yongin Mega Cluster (1) (2), six fabs capable of 2nm and below, part of a $471 billion USD national project with Samsung and other firms, first phase operational by 2030. The goal is 7.7 million wafers per month starting in 2030.

It's important to understand that chip production is not really a pure private endeavor. For mega-scale chip manufacturing to work, it's always been a potent combination of government and private industry. Governments often invest massively, clear regulatory hurdles and give sweetheart tax breaks to own this lucrative industry. This is true in both South Korea and Taiwan and throughout Asia, a style of business that's not natural to US industry outside of the 1950s when government was more tightly in step with business after the war.

Take TSMC and its origin story for instance: "Alongside generous tax benefits, the Taiwanese government, through the National Development Fund, Executive Yuan, provided another 48 percent of the startup capital for TSMC, and the rest of the capital was raised from several of the island's wealthiest families, who owned firms that specialized in plastics, textiles, and chemicals. These wealthy Taiwanese were directly 'asked' by the government to invest. 'What generally happened was that one of the ministers in the government would call a businessman in Taiwan,' Chang explained, 'to get him to invest.' From day one, TSMC was not really a private business: it was a project of the Taiwanese state."

There's more. When "K.T. Li (李國鼎), known as the “father of Taiwan’s economic miracle,” convinced Taiwanese-American Morris Chang (張忠謀) to move back to Taiwan and help build up Taiwan’s semiconductor industry. When Chang founded TSMC in 1987, he did so with $100 million in seed money from the Taiwanese government."(1)

TSMC benefits from massive tax credits and multi-year tax reduction deals. The same is true in Korea, where the new chip clusters will benefit from cleared hurdles for power plants and tax benefits for R&D and more. Expect the effective tax rate of the facilities to be roughly 0% for many years when all the numbers are actually crunched and all the credits and building costs are applied.

So we have very little capacity in the US, a new push by Korea to match Taiwan in five years and nothing worth noting anywhere else on the planet.

And when it comes to US facilities, TSMC is also keeping the most advanced and most cutting edge next-gen factories in Taiwan.

Why?

The "silicon shield".

Silicon and Lead Shields

The Silicon Shield is a simple idea. It means that if China invades they won't get access to the best chips anymore so they won't invade. It also means that world supply chains would be so disrupted and destroyed that the rest of the world would be forced to fight. More on that later.

"Following the explosion of the semiconductor industry in 2020, then-Taiwanese President Tsai Ing-wen (蔡英文) penned an op-ed in Foreign Affairs in which she wrote, “Our semiconductor industry is especially significant: a ‘silicon shield’ that allows Taiwan to protect itself and others from aggressive attempts by authoritarian regimes to disrupt global supply chains.”"(1)

One part of the silicon shield is already crumbling.

China is not getting access to the best chips anymore, since the US pressured TSMC to cut them off, part of an increasingly aggressive campaign to put a ringed fence around China's development.

In Nov 2024, the US Department of Commerce ordered TSMC to halt shipments of 7nm and more advanced chips (including 4nm, 3nm, and 2nm) to Chinese customers, effective immediately. That means local Chinese companies can't design their own advanced chips and get them made in Taiwan like the rest of the world.

It's a measure of just how important US tech giants are to TSMC business that they complied with such a foreign order in the first place. Without American tech revenue and American purchasing power, TSMC doesn't really exist. Without TSMC, Apple, Nvidia, Qualcomm don't really exist.

So far this fantastic, global division of labor has held, an international unwritten pact that constitutes the most advanced and complicated supply chain in the history of the world. If you don't like globalization, you don't like the smart phone in your pocket or the laptop on your desk or the TV on your wall. These things are impossible without a massive, interconnected planet.

But at the moment it means we're in a very bad state where the US doesn't have nearly enough manufacturing capability and no plans to build it at the moment. The US passed the Chips and Science Act in 2022 which pumped roughly 59B USD into the push to get fabs on US shores.

That amount of money is shockingly small for the US and it just won't be enough. Not even close. It doesn't even match the original seed money of TSMC in the 1980s or the 471B USD that Korea will spend on juicing up their own manufacturing over the next five years.

In fact, it's a pathetic amount by comparison.

Even worse, we can't even seem to commit to that kind of chump change, much less more. The incoming commerce secretary doesn't seem to understand the strategic significance of the Chips Act and offered only tepid support. This kind of vapid, wavering support on the part of the US is not going to get any significant manufacturing capability on shore any time soon.

As we've seen from Taiwan and Samsung, it requires strong, unwavering partnership between government and private industry over many, many years, the kind of investment and commitment we just can't seem to understand anymore.

The Chips Act and our pathetic and wavering support for it is not nearly enough to get the capacity we'd need to avoid a total economic meltdown on Chinese invasion on Taiwan.

And I mean total.

Think every major car manufacturer absolutely ground to a halt and no more new computers or AI datacenters. What are you going to fill the 100B USD OpenAI and Softbank Stargate facility with if there are no chips? When we had massive car and truck manufacturing slowdowns during COVID, it wasn't because they couldn't make the steel and carbon fiber frames of the SUVs, it's because they ran into a global chip shortage as the factories in Taiwan faced disruption after disruption. That was just a small taste of what would happen following an invasion.

Now is there any way to fix this complete mess?

Recently we saw a push by Sam Altman, as reported by the Wall Street Journal, that he was looking for as much as 7 trillion to rebuild the chip pipeline across the world. In the mass media that was met was eye rolls and scorn, but I'd argue that was mostly out of sheer and utter ignorance at the scope of the problem and how precarious the modern world rests on that supply line. That kind of number is exactly the kind of might and muscle it would take to reshape the global chip supply line.

The incredible 2022 book Chip War, by Chris Millar, goes into the costs and global nature of the chip supply line. A few numbers will help put it all in perspective for you.

A new cutting edge chip factory costs roughly 20B to build. That’s billion with a B!

ASML, a dutch company, makes 100% of the EUV (Extreme Ultraviolet Lithography) equipment that make the most advanced chips possible. It's old EUV machines, so-called Low-NA EUV, cost $180M USD per machine. It's latest, High-NA extreme ultraviolet (EUV) machines cost $380M USD per machine.

Those machines themselves are a modern miracle of engineering and integration. They’re powered by a series of parts from one-of-a-kind companies, each of which is essential to making the complete machine.

"ASML itself only produced 15 percent of an EUV tool’s components, he estimated, buying the rest from other firms. This let it access the world’s most finely engineered goods, but it also required constant surveillance. The company had no choice but to rely on a single source for the key components of an EUV system. To manage this, ASML drilled down into its suppliers’ suppliers to understand the risks. ASML rewarded certain suppliers with investment, like the $1 billion it paid Zeiss in 2016 to fund that company’s R&D process. It held all of them, however, to exacting standards. “If you don’t behave, we’re going to buy you,” ASML’s CEO Peter Wennink told one supplier. It wasn’t a joke: ASML ended up buying several suppliers, including Cymer, after concluding it could better manage them itself," writes Miller in Chip War.

Zooming in a little we see just how complicated and intricate this supply chain is and why replacing it with 7 trillion dollars is not an absurd number.

Cymer, the San Diego company ASML bought, was originally founded by some laser experts, and they made the key EUV light sources with lasers and tiny balls of tin blasted at half a million degrees, many times hotter than the surface of the sun. But they could only make their laser because of a German precision tooling company Trumpf, which makes a carbon dioxide based laser that could meet Cymer's specs.

"It took Trumpf a decade to master these challenges and produce lasers with sufficient power and reliability. Each one required exactly 457,329 component parts,' writes Miller.

These are fiendishly complicated machines and manufacturing pipelines, built on a string of companies that cannot be replaced and that are scattered throughout the world.

What is the US doing to deal with the problem outside of pumping a measly 59B into semiconductors and waffling on its commitments?

Trump recently proposed tariffs of 25% to 100% on Taiwan made chips. It's the president's usual type of gambit. He typically takes an extreme position to terrify people and then walks it back to something more manageable in order to get concessions and announce a deal. In this case he hopes the concessions will be more chips on US shores without further mega-packages like the Chips Act. The president also believes "trade wars are easy to win" a view that doesn't match history at all, in the least.

This gambit won't work and it will drive up prices for US consumers. And even if the bluff kinda-sorta works, it just won't be enough to get 12 or 20 factories of 20B a piece onto American shores any time soon. Actually carrying through on the tariffs would be an even bigger disaster, driving up the prices of components across the entire American industrial pipeline with no feasible alternative anywhere in sight.

So the short answer is we have nothing.

Meanwhile, Xi Jinping of China said in his New Year's speech, "The people on both sides of the Taiwan Strait are one family. No one can sever our family bonds, and no one can stop the historical trend of national reunification."

China ratcheted up pressure on the island in recent years, building specialized landing boats, doing military exercises around the island and more. It's not a matter of if they will invade, but when.

In a 2022 article in Time, Chris Miller writes that it won't be easy for China to simply take over TSMC facilities:

"They’d soon discover that crucial materials and software updates for irreplaceable tools must be acquired from the U.S., Japan, and other countries. Moreover, if China were to invade, it’s unlikely to capture all TSMC employees. If China did, it would only take a handful of angry engineers to sabotage the entire operation. The PLA’s proven it can seize Himalayan peaks from India on the two countries’ disputed border, but grabbing the world’s most complex factories, full of the world’s most precise machinery—that’s a different matter entirely."

It need not be a complete invasion. China may simply launch of series of cyberattacks, or missiles to destroy and disable Taiwan's military capabilities, or blockade the island. When a country can't get food and supplies in, it can break down quickly. They might simply demand access back to 7nm and below chips again and Taiwan and the world would have few options, either global world war or acquiesce to demands.

And of course, sometimes rationality just don't play a part in nation state moves.

Trade Wars and Why Nations Sometimes Do Stupid Things

In the lead-up to both World War I and World War II, one of the most prevalent beliefs was that the world had become too economically interconnected for a global conflict to take place. Prior to World War I, the world experienced a period of significant globalization and economic integration that looks laughable compared to the interconnectedness we have now but was still foolishly considered a deterrent. Norman Angell's influential 1933 book "The Great Illusion," argued that "the economic cost of war was so great that no one could possibly hope to gain by starting a war the consequences of which would be so disastrous." For that reason, a general European war was very unlikely to start, and if it did, it would not last long. He argued that war was economically and socially irrational and that war between industrial countries was futile because conquest did not pay.

You know how that turned out. World War II.

Of course, that's just the thing, people are not rational. Neither is war. Neither are realpolitik and power plays and conquest and human delusion and megalomania and ultra-nationalism.

Miller is a fantastic thinker and Chip War is a tremendous book but in his article he puts too much emphasis on logical thinking in nation state decision making.

There is sometimes nothing rational about it.

Sometimes governments just do stupid things, just like people do stupid things. Again, at the outset of WWI people though that the world was just too interconnected to risk an all out world war. They got proved wrong the hard way b.ecause people aren't guided by rational, clear thinking. They're driven by base instinct.

When it comes to Taiwan, rational, clear thinking won't play a part.

If the US continues to insist on ratcheting up chip controls, of the kind that Anthropic CEO Dario Amodei is pushing for to eliminate Chinese competition to his foundation model company, and that the Biden administration rushed out the door at the zero hour with no debate, an executive order which turns the world into fiefdom's of colonial chip export controls and puts many of our allies as second class citizens (like Poland, Portugal, Switzerland and other headscratchers) with very little rational, then it's likely to push China to do the irrational even faster. The Trump administration has done nothing to change the export controls and it seems the incoming commerce secretary is hungry for more. This is disastrous thinking on every level that puts the entire world at risk.

(Source: Bloomberg)

China is also racing to recreate the global chip manufacturing pipeline itself. It's struggling. It's not just a matter of spending but of owning key technologies.

Still, given enough time, will power, and government plus private industry commitment, the kind that Asia excels at the the US doesn't, it's likely they will succeed and then they can invade just to disrupt the supply lines for everyone else. An alternative breakthrough to EUV, the kind of which TSMC are already hinting at, could change the game for everyone.

Don't count China's tech companies out.

There's a sense from the most jingoistic and idiotic people in the US that Chinese companies just copy American technology. These are usually folks who've never set foot outside of the US and have certainly never been to Asia. Their idea of China is still stuck in the 1980s with a bunch of people on peddle bikes in Beijing. That China is long gone.

(Source: Shutterstock, China in the 1980s)

Today they drive electric scooters, have the number one car manufacturing industry on the planet, along with advanced robots that can walk and run like people, self driving cars and more. In just twenty years China has gone from producing almost no cars to producing 30 million vehicles a year, double what the US produces and at 20% to 25% of the cost of the rest of the world.

(Source: BoFa Global Research and Bloomberg)

People in the US who think China is just a low-rent copy cat economy are dangerous to America because they're pushing a picture of a China that doesn’t match reality in the least.

In an invasion scenario, we'd better hope that Korea has successfully upped its capacity capabilities by a huge margin or that Altman or someone else ambitious got those trillions of dollars to remake the chip pipeline because we'll wake to a world where all the chips that have already been made are all the chips that are getting made for many years as we scramble to try to build out 20B factories in record time.

It's a scenario where even the Trump administration wouldn't be able to keep us out of World War III. The US would be forced to defend Taiwan and so would Europe and much of the world or risk total economic meltdown.

That's not a world I want to live in.

To avoid it, we better hope we have a better answer than "trade wars are easy to win" by the time it happens.

Sam Altman and Dario Amodei, CEOs of OpenAI and Anthropic, the top two frontier AI labs in the world, recently published their vision for a sparklingly hopeful AI future with The Intelligence Age and Machines of Loving Grace.  There's much to love in both of these essays and they were much needed because the narrative around AI has turned positively poisonous over the past few months.

We've seen a status game of ever-more unhinged predictions about the end of the world, amplified by a mass media which thrives on fear-driven clicks, and pushed by a billionaire backed Doomsday cult doing an all out media blitz for radical legal intervention with California's SB1047, which thankfully failed, though other attempts to strangle AI development are on the horizon with more than 700 convoluted, overlapping bills advancing at the state level.

But the biggest problem is not whether we need more utopias or dystopias. 

What we most desperately need is a heavy dose of realism. 

We need the middle path of AI. 

Unfortunately, nuance and reality don't sell very well. You win more friends, money and social media followers screaming extreme ideas about heaven and hell.

We're all going to going to live forever, dancing into the sunlight in a bold new age, and they'll be no more disease and no more suffering and no more pain!  Or, if we don't act right now we're all going to dieeeeeeeeeeeeeee!

To get a realistic picture of AI we need to concentrate on seeing through the fog and the noise. The middle path means focusing on the most likely futures, the ones at the meaty center of the normal distribution. 

The Utopian future fantasies of the Culture sci-fi series cited by Amodei, versus the runaway, recursive AI self-improvement of Yudowski, aka AI FOOM, are binary, extreme visions that make for great stories but they're not a great reflection of actual reality.  If we're always focused on heaven or hell, we have no hope of getting any kind of clear picture of how AI will actually impact life, work and the world.  Even worse, we have no hope of building useful legal frameworks or strong guardrails and of letting regular folks just go about their business without jumping through useless hoops because we've shackled all these systems with a suffocating bureaucracy. 

This kind of black and white, all-or-nothing, binary thinking warps and distorts reality.  It's a dark, oily smoke that covers everything and makes it impossible to see what's actually right in front of us.  We can't get a clear signal because it's drowned out by the constant screeching.

The big story leading up to the election was how AI driven misinformation would overwhelm our ability to tell reality from fantasy.  Turns out it was yet another phantasm of the imagination and we learned once again that politicians didn't need any help creating misinformation.  They've been doing it just fine for tens of thousands of years.

The imagined problems from extreme predictions rarely play out. Extremes are the edge cases on any probability distribution of the future. 

Most often life is somewhere in between. 

History may seem like one big event after another but it's only because you can't tell a good story without conflict. Stories are conflict.  They don't exist without them. That's why we rarely see history how it really is: long periods of peace, punctuated by short periods of war. 

While we do have periods of extreme darkness, like World War II, or brilliant light, like the pastoral pre-colonial days of the Americas, they're usually short intervals in the grand flow of time, sharp spikes before life inevitably returns to a mundane, every-day reality that's somewhere in between. 

It's easy to think of dark societies, like Russia under Stalin or Germany under the Nazis or China under Mao.  These societies have one thing in common almost inevitably:

They're self-created.

Notice how these tend to be societies ruled by a single psychotic personality cult that creates a nightmare for a time until that person dies and gets replaced by something much less extreme.  They're created by the very people prone to extreme binary thinking: "If we don't take over and impose communism on everyone and kill everyone who disagrees the world will end!" 

In a horrifying irony, it's extremist thinkers who create the worst histories. 

They create the very things they fear through their paranoia and delusions or their desire to achieve a "perfect" world or to avoid a great "apocalypse", one that usually goes through a river of blood created by the extremists.

In the same way, some societies are immensely positive for the vast majority of their people.  We can think of early native American societies or the vast nation states of America and Europe and even China today, with far more wealth, prosperity and technology than it ever had in its 5000 year history. 

Even if China today lacks in political freedom and free speech, it's still a world away from the insane era of Mao that saw 100s of millions of Chinese die from starvation, war and political violence in the Great Leap Forward and the Cultural Revolution, or the time of warlords pre-Mao, where a few hundred people had the power of life and death over nearly a billion people and could hang any peasant for stealing a loaf of bread to feed their families.

If we're clouded by these fantasy fears and utopian dreams we're virtually guaranteed to miss the most likely futures, one where AI does some good and some bad but it exists on a sliding scale that's tilted closer to good or closer to bad depending on our choices but where AI is not "catastrophic" or curing all diseases overnight and solving the Riemann Hypothesis in its spare time. 

Here's the truth: No matter what we do, AI will do good and bad things.

Let's be even more clear:

We will not stop all the bad uses of AI and guarantee that AI only gets used for good.

That's impossible. 

No matter what policies we set, no matter what kind of government we choose, life always has positives and negatives.  There is no society in history, ever, and there never will be, that is all positive or all negative.  That is for children's books and great, door stopper fantasy epics.  Instead it's best to look at societies and cultures on a continuum of light to dark at the same time. 

The question we have to ask is:

Does a society, on the whole, produce more positive outcomes and lives for its people or more negative ones?

To get more of the good and less of the bad, we have to ask ourselves what kinds of policies are likely to lead towards less negative uses and more beneficial uses of AI?

To understand how to get more of the good things and less of the bad things, we have to see clearly.  We have to set aside Heaven and Hell and focus on the more likely parts of the probability distribution of the future so we can make practical plans and shepherd ourselves into a better tomorrow.

So what does a realistic picture of the future look like?

The World of Tomorrow

To start with, AI is not coming, it's already here and it's already radically changing the world.

ChatGPT, the "technology nobody wanted" to its detractors is the the fastest growing software app in the history of the world, rocketing to 100M users in just two months. It took TikTok nine months to hit the same marker and Instagram three and half years. ChatBot usage is estimated at over 500M users worldwide in the two years since ChatGPT launched (Nov 2022) on the direct consumer guide, and if you factor in service chatbots on websites, its billions. 

AI is everywhere already.

AI is already driving cars, beating people at ancient board games like Go, pummeling people in complex tournament video games like DOTA 2, and detecting diabetes.

(Source: Valve)

When you talk to your phone and it understands you want directions to a Thai food restaurant, that’s AI. We just don't think of it as AI anymore because it just works. 

My direct use of Google dropped 80% in the last six months because I now turn to Perplexity an AI intermediary to search.  In its early days it hallucinated wildly but now it lasers in on near perfect answers almost every time. It's faster for me to do research and get exactly the answers I want without wading through pages and pages of irrelevant text and ads trying to answer one simple question about what wax tablets were called in ancient Greece.

(It's called a deltos.)

Machines hunt down terrorist’s money and Taylor Swift used facial recognition to keep her safe from psychotic predators at concerts. She’s not the only one. As early as 2018, long before ChatGPT rocketed AI into the public consciousness, Chinese police were using it to track criminals and dissidents.  "Chinese police arrested a 31-year-old suspect who was hiding among nearly 60,000 people during a concert at Nanchang International Sports Center. This was made possible by the early stages of China’s own 'Xue Liang,' or 'Sharp Eyes,' monitoring system, which will be tasked with monitoring the movements of its citizens."

Facial recognition is rolling out to airports everywhere all over the world and it’s coming to a street corner near you.  The last few years when I roll in from overseas with Global Entry the system just takes my picture and that's it. I don't even need to scan my passport.

Meanwhile, Alphafold has already predicted over 200 million proteins and then moved on to understanding how they connect or "dock" with each others along with open source mirroring their approach, like ESMFold from Meta. That's a catalogue of almost every "protein known to science, including those in humans, plants, bacteria, animals and other organisms" writes the Washington Post in their article on the three people who just won the Noble prize in chemistry. Two of them, Demis Hassabis and John Jumper of DeepMind, led the effort on AlphaFold. 

The article goes further.  This year’s Nobel Prize in chemistry honored "a real-world example of how AI is helping humans today with astounding discoveries in protein structure that have far-reaching applications".

Right now AI is a mustard seed.

But from those seeds will grow a wild forest that ripples through every aspect of life from top to bottom.

So that's where we are.  But where are we going?

Back to Reality

First off, we likely won't get AGI (Artificial General Intelligence) in two years, like Dario promises but that's just fine. I've written before about what's missing to get to AGI and how we might get there in my article Why LLMs are Smarter Than You But Dumber Than Your Cat.  The short answer is new algorithms and new breakthroughs are needed and I don't see them on the horizon just yet. 

"Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, told Reuters in a recent article that results from scaling up pre-training - the phase of training an AI model that use s a vast amount of unlabeled data to understand language patterns and structures - have plateaued."

'The 2010s were the age of scaling, now we're back in the age of wonder and discovery once again. Everyone is looking for the next thing,' Sutskever said. 'Scaling the right thing matters more now than ever.'"

But that's just fine. 

What we will get is still incredible and that's rapid growth in ambient intelligence (more on that in a moment) over the next few decades and beyond.  And we don't even need AGI to get there.

Specialized intelligence is amazing too.

We tend to think of human like, ultra flexible intelligence as the ultimate goal, but we're just another kind of specialized intelligence.  By being good at lots of things we're jacks of all trade, masters of none.  Generalized intelligence is, weirdly enough, a kind of specialized intelligence in itself, one that has big trade-offs.  Beyond gorillas, monkey, people, octopus, dolphins and a few other creatures, generalized intelligence is in the minority.

Instead, nature abounds with specialized minds.

You might not think of a squirrel as intelligent, but try to keep that squirrel out of your garden and you'll find an intelligence that's hyper-tuned to get into your garden.

And that's where we're going for sure.  Hyper-specific, hyper-tuned AI that's perfectly suited to the task at hand as the vast majority of AI systems.

Welcome to the age of ambient intelligence, with embedded intelligence infusing every aspect of our daily lives.

It will track shipping containers around the world in realtime, looking down from on high through the eyes of satellites. It will drive cars, track your health, help your doctors figure out what's wrong with you when you're sick, spot tumors before they can grow, do science and material discovery faster, design better and faster chips and more. 

Once dumb objects will wake up. Your shirt will babble away with your shades while having a conversation with your girlfriend’s pearl earrings when she’s traveling to a vacation in Brazil.

Everything from our houses, to weapons, to planes, trains and automobiles, to roads, clothes, jewelry, headphones, glasses, and eye contacts will grow wild with thoughts.

Deep learning, symbolic reasoning and genetic algorithms will look primitive by comparison to the rapid learning systems of the 2030s to the 2040s, with hardware based neural networks that run at ridiculously low power, with tightly coupled memory and processing in a single pass running on memristors and light based chips and organic wetware chips.

Our smart bandages will monitor our wounds and tell us when to change the dressing or whether the infection is healing or getting worse. Tiny patches in the bandages will react to infection by releasing antibiotics.

Pill bottles will remind us to take our medicine or warn us not to take it if we wake up groggy from a night of drinking and try to take it twice.

Self-driving trucks with no cab for humans will race along the highway at 200 miles per hour on special roads across China and the US. They’re nothing but stripped down flatbeds with wheels, carrying standard smart shipping containers. The containers will know every single object inside of them, the objects scanned by nanosats and tracked in space forever on the AR/IoT web. As soon as dishes come off the industrial 3D printer at the autofactory, they’re tracked right into your house.

(Source: Logan movie: Self-Driving trucks in a surprisingly subtle and amazing take on the future)

The containers will arrive on self-piloting boats. When pirates try to hijack the containers, the boats will go to evasive maneuvers and the containers will screech and call for the coast guard or the navy or they'll unleash a torrent of deterrent drones that swarm and crash down on the pirates to sink their boats.

Your next black leather jacket in 2030 will hit your house via drone delivery. When you go to Berghain in Berlin to dance the day away, your jacket will adjust its threads as the temperature changes or loosen the threads as it gets hotter. No more wearing layers as you move from hot to cold in the ice and snow of winter to the warmth of a self-driving car.

We already have Meta smart glasses that just look like regular shades.  They're already light years away from the hideous Google Glass of a decade ago.  This is the worst they will ever be.  They will only get better from here. 

(Source: Ray Ban shop and Meta)

Our smart glasses will evolve into designer contacts and eventually designer prosthetic eyeballs, that show us a constant overlay of meta-information about the world around us, from the contact information of people we know, to details of historical landmarks, to warnings of high crime areas to avoid, to the location of friends and family who have given us permission to see where they are during certain times of the day.

(Source: Psycho-Pass hologram AI assistant.)

An AI personal assistant will split time between your ear buds and your augmented reality contacts, processing half of its routines on distributed cloud and foglet platforms. It will do all kinds of tasks that used to take hours. It will bug you to get up in the morning, set meetings, find a nice bottle of wine on the way to a party, craft an awesome workout to get you into summer shape, suggests the latest fashions and finds the perfect restaurant to take a date, while telling you what to say on that date to help you seal the deal with a night of passion. 

It'll keep your secrets like Di-Di, Judy Jetson's floating digital diary in the Jetson's, and act as a trusted friend and advisor.

(Source: The Jetsons)

Kids used to have to use their imagination with their toys. Remember army men and action figures? You had to pose them yourself and give them voices and stage epic war sagas with them in the backyard.

Tomorrow's kids won't want a teddy bear that just sits there slack-jawed, staring at the wall. Teddy bears will walk and talk and get kids to beg their parents to buy things. And action figures that don’t run and leap and charge the backyard battlefield screaming will be worthless to tomorrow's children except to collectors who pine for simpler times when the mind’s eye had a lot more work to do.

Toys used to say “batteries not included.”  Now they'll say “no imagination needed.”

It’s Toy Story come to life.

(Source: Teddy from the underrated movie A.I.)

Bikes will have wobble detection and bizarre alien frames that look strange because were designed by machines to make them more stable. You could probably peddle them if you wanted to, but if you get tired the AI coupled with millimeter precise GPS takes over and drives for you.

Fire suppression systems won’t just put out fires in houses, they'll direct families where to run and what rooms to avoid, all while keeping them calm with a soothing voice and calling autonomous fire trucks.

But let's look closely at some of the most powerful uses cases for AI.

Dario Amodei covered one of them beautifully in Machines of Loving Grace:

Health care and biology.

Health Care and Biotech Boom

Amodei rightly covers many of the limiting factors to dramatically improved science and biotech:

"In the “limiting factors” language of the previous section, the main challenges with directly applying intelligence to biology are data, the speed of the physical world, and intrinsic complexity (in fact, all three are related to each other). Human constraints also play a role at a later stage, when clinical trials are involved."

He writes that "we could get up to 1000 years of progress in 5-10 years":

"Thus, it’s my guess that powerful AI could at least 10x the rate of these discoveries, giving us the next 50-100 years of biological progress in 5-10 years.¹⁴ Why not 100x? Perhaps it is possible, but here both serial dependence and experiment times become important: getting 100 years of progress in 1 year requires a lot of things to go right the first time, including animal experiments and things like designing microscopes or expensive lab facilities. I’m actually open to the (perhaps absurd-sounding) idea that we could get 1000 years of progress in 5-10 years, but very skeptical that we can get 100 years in 1 year. Another way to put it is I think there’s an unavoidable constant delay: experiments and hardware design have a certain “latency” and need to be iterated upon a certain “irreducible” number of times in order to learn things that can’t be deduced logically. But massive parallelism may be possible on top of that¹⁵."

That's possible but the more likely probability is that we do gets lots of progress in many areas, but we run into one last limitation that he seems to miss:

Our bodies are insanely, mind-bogglingly complex.

The level of complexity at the molecular level is utterly astounding.  Check out this rendered video below on just one of the fiendishly complex molecular machines working inside your body right now:

Heliocase, spins as fast as a jet engine and it winds at unwinds DNA molecules and astonishing speed, each one of them "a few centimeters long but just a couple nanometers wide."

What kind of scale are we talking about?  How big is a nanometer?

"Two nanometers is about 1/50,000th the width of a human hair.  A human hair is approximately 80,000-100,000 nanometers wide."

(Source: Visual Capitalist)

Those little machines, clustered together, can rip apart and replicate your DNA with ridiculous levels of accuracy in about 8 hours.

We have about 30-40 trillion cells inside of us, all with DNA inside of them and we have billions or even trillions of just this one molecular machine doing this one incredibly complex job.

The optimistic view and the doomer view that an AI explosion will take off and unstoppable speed, dramatically underestimate the complexity of the biological world.

The idea that we will cure all diseases in a decade is quite frankly, absurd.

But that doesn't mean we won't have incredible breakthroughs.

There's nothing but intelligence that can help unravel the micro-universe inside our bodies. We don't have enough scientists and enough time and enough money to do it fast enough.  It doesn't matter if we see a 5x, 10X (or 100X speed acceleration of Dario) in our biological understanding and experimentation, or more modest gains somewhere in between those wild-eyed predictions.  No matter what, AI will accelerate science and healthcare. 

We’ll point our smart phones at a worrying spot on our arm and the AI that lives in our phone will tell us to call the doctor. When it sends the report it generated to the doctor’s office, the triage nurse will know to bump us to the head of the schedule if its skin cancer or a festering wound, rather than treating every patient as equal from hysterical hypochondriacs to the old lady who just likes talking to doctors.

Just this year the Sybil system out of MIT was able to accurately predict lung cancer risk up to six years in advance.  I take such prediction systems with a grain of salt, but they will get better over time.  Imidex has a lung cancer system that's already FDA approved that can detect cancer cells up to 6nm with 83% accuracy.  Again, this is the worst it will ever be.

Big improvements in health care mean dramatic drops is health care costs too.  That's going to be utterly essential as population sizes drop in the coming decades and less people are paying into the health care systems.  Right now, false positives and negatives mean patients get the wrong care or get care too late or worse don’t get care at all. Big improvements mean cheaper health care and people living a lot longer because they get the care they need when they need it.

The biggest hurdle to AI in health care is the overly conservative and overly regulated nature of the system in almost every country in the world, whether that’s the cold blooded big money system of the US or the sweeping socialized systems of Europe.

It’s hard to get new devices approved. The barrier to entry is massive. Small boards of people we don’t know act as a choke point, both protecting people from hucksters and slowing down innovation dramatically with their double edged sword.

Expect those walls to start to come down as AI increasingly detects and cures disease and performs research at scale.  People will demand access to those breakthroughs and regulations will crumble.  Laws will change to make it easier for AI based health devices to get into hospitals and into our phones and augmented reality glasses.

As the walls come down it will open up the possibilities for preventative care at a more personal level. Your watch and your glasses will know that your frequent napping during the day isn’t just overwork, it’s sleep apnea.

That will streamline the convoluted process we have now where you make an appointment to get a sleep study, then go to another appointment to get trained for the sleep study, then take the machine home and do the sleep study, then take it back to the office to get your results, and then see a doctor in yet another appointment to prescribe the actual CPAP machine that alleviates the problem. 

All this to get an air machine.  It's basically a scam.

Archaic and convoluted processes like this will start to fall to the power of AI because you will walk into the doctor with proof that you already have sleep apnea.

Hospital devices will get smarter too, from blood pressure machines, to cameras, to dynamically adaptive physical therapy machines. Cameras will detect if old folks stop breathing, have a heart attack, or take a nasty spill.

All of our devices will start to work as preventative care devices or more Tricorder like universal medical devices, bundled together and able to detect many different things in a single unit.  No it won't just magically detect everything with perfect accuracy, but we'll increasingly have medical capabilities in our wearables and especially in any internal devices like the neuralink or small devices we can embed under our skin that talk to our devices.

Your watch or ring or contact lens will increasingly become your heart and health monitor. In ten to fifteen years it may even be able to predict a heart attack coming before you go face down in your eggs and bacon.

“ALERT: Please go to an emergency room immediately as you’re in danger of a heart attack.”

That will mean you get to the hospital in time to live a lot longer.

Ambient intelligence will help us scale up health and science in incredible new ways, as machines invent new drugs, do science experiments with robo-labs, tackle Alzheimers, crush many kinds of cancer and maybe even unravel the basic mechanisms behind cancer and give us near-universal cures.

So in many ways I'd love to see Dario's version of the future become reality, but I'd bet that the incredible complexity of the body will remain a difficult mountain to climb and we'll see incredible progress but many diseases will continue to remain elusively out of reach or new diseases will evolve as we live longer lives and better lives.

Entertainment and Games

It's 2035 and you're a top notch concept artist in London building a game with 20 other people.

The game looks incredible. It's powered by Unreal Engine 9 and it's capable of photorealistic graphics in real time on the Playstation 7 and near perfect physics.  It's got character AI that can talk with anyone in an open ended way and still stay within the plot guidelines to keep players hunting down magical rubies and the divine sword of radiance. 

Ten years ago it would have taken a team of 1500-5000 to make this giant game.  Now you can do it with 20 people.  But that doesn't mean less work and less games it means more.  A lot more.  We used to get 10 AAA games a year and now we get 10,000.   Even with a 1000 folks you probably couldn't do it back in 2024.  That's because it's almost completely open ended, with real time procedural storytelling that creates nearly infinite side missions, personalized music that matches the listener's tastes, and players control it with hand gestures and voice commands.  The controller is gone. 

AI helped write the code for UE9 game engine, helping Unreal programmers do four releases in ten years, instead of a single release every decade.  AI refined much of the original game engine code and made it smaller and more compact, while auto-documenting it in rich detail.  AI also helped design the Playstation 7.  Chips design models crafted layout of the transistors on the neural engine and graphics chip inside the Playstation.  But where AI really accelerated the game development process was in design and asset creation workflows.

As the lead artist, you've already created a number of sketches with a unique style for the game, a cross between photorealism and an oil painting aesthetic that mirrors classic, clean, golden age sci-fi.  The executive creative director signed off on it as the official look of the game.  You fed the style to the fine tuner and the AI model can now rapid craft assets that share that look and feel. 

Art isn't dead.  It's become co-collaborative.

Nothing can replace the human artistic spirit but it now works hand in hand with state of the art generative AI in a deeply co-creative effort.

Now you can move much faster.  You quickly paint an under sketch of a battle robot sidekick and the tough female soldier for the game plus a bunch of variants that you were dreaming about last night.  You feed it into your artist's workflow and talk to it and tell it you want to see 100 iterations in 20 different variations.  It pops them out in a few seconds and you swipe through them. 

“I’ve changed my mind," you say.  "Go back and give me 200 iterations."

The natural language and logic model baked into the app understands what you want and does it instantly.

300 seconds later the high rez iterations are ready.  The 27th one looks promising so you snatch it out of the workflow and open Concept Painter and quickly add some flourishes: a different face mask that generative engine didn't quite get right; a new wrist rocket.  You erase the laser shoulder cannon and replace it with an energy rifle and then you feed it back to the engine.  You level up the organic, biomechanical nature of the armor, making it drip pure biopunk.

It pops out 50 more versions and you pick the 7th because it's nearly perfect and you just need to fix a few things.  You rapidly shift it to match the style of the game assets with another step in the AI pipeline and then you check it for artifacts.  You fix a few problems but now it's good to go, so you send it off to the automatic 3D transform. 

It pops out into the 3D artist's workflow, fully formed just a few minutes later.  The 3D artist, working from Chiang Mai, Thailand, has to fix a few mangled fingers some clothing bits that don't flow right and then it's ready to go.  He kicks it off to the story writer working out of Poland.

Seeing the new character gets her inspired.  She quickly knocks out a story outline in meta-story language and feeds it to her story iterator.  It generates 50 completed versions of the story in a few seconds.  She starts reading.  The first three are crap but the fourth one catches her and she finds herself hooked to the end.  It's good but it needs a little work with one of the sub-characters so she rewrites the middle and weaves in a love story with that character and then feeds it back to the engine for clean up.  The new draft comes back and it reads well so she fires it off to the animator in New York City.

He pulls up the draft and the character and tells the engine to start animating the scene.  In five minutes he's got story stills for the major scenes in the story and branching possibilities for character actions the player might take.  Most of them are good but the third scene is off, so he does some manual tweaking and then asks the engine to iterate again.  This time they come back good and he sends it all off the engine to complete the animations and connective tissue linking the scenes. 

Welcome to the Age of Industrialized AI. 

Even students and home film enthusiasts are making games. The Quentin Tarantino's of tomorrow will use AI to get out of their crappy day job and go on to become great directors.  They'll still need incredible skills, like knowledge of story, and character and development.  They'll need to know lighting and how emotions come across on the screen.  They'll need to understand camera movements, even if those cameras a virtual. 

The people best poised to benefit from AI are the professionals who already know the real world of film and TV.

Those students and directors and professionals will have a vast dashboard of new tools. They will animate people and fully realistic scenes the way we used to animated cartoons. They will give subtle expressions and performances to the AI actors, and incorporate real actors seamlessly into the scene.  They will "reshoot" a scene in minutes and create a thousand variations of it. 

The kids that grew up with AI won't know the bitter controversy around it in the arts and they won’t care if you told them about it.  To them, it will just be another tool, like a hammer or a phone and they will use it to build incredible student projects, many that go on to great acclaim and vast audiences.

Expect an absolute explosion of new film and TV and games that adapt to you on the fly, a golden age of story.

Agents, Robots and the Future of Work

But what about work? Will robots and agents take all the jobs?

No. But people working hand in hand with robots and agents will absolutely overtake anyone who refuses to work with them. The AI won't replace the radiologist but the radiologist working with AI will replace the one who won't use it.

What are agents?

My team at Kentauros AI defines it simple as:

An AI system that's capable of doing complex, open ended tasks in the real world.

Even more complicated?

Build an agent that can do long-running, open ended tasks.  They need to work for hours, days or weeks without any unrecoverable errors.  Just like a person.

The real world could be the physical world or online.  What we're talking about here is when agents get out of the lab and face the never ending complexity of life.

Of course, life is messy.  When you deal with the chaos and complexity of the real world, it's incredible hard to make these systems work well.  Most teams are badly underestimating just how difficult it will be and throwing out wildly optimistic predictions of how fast we'll have robots and agents in every day life.

Moravec's paradox says that reasoning for robots is relatively easy and sensorimotor perception is hard.  That's wrong. Both are hard, especially when you try to generalize reasoning. 

We've seen Tesla cars go from software based object detection and perception to built in hardware based object detection on specialized chips.  As we understand more and more of how to make machines understand the world, we'll see more of that capability in silicon.  Steve Jobs once said that "software was everything we didn't have time to get into hardware." 

Right now we might have to spend a ton of inference time compute on "hearing" but your ears are fine grained hardware based fast Fourier transformers. They do the work in an instant in hardware.  Expect robotics to follow more of the same, with a flurry of specialized chips and devices that do the work of seeing and interacting with the world.

If you're paying attention to robotics work, we seem to have made more progress on sensorimotor perception over reasoning so far.

We have a joke in house:

If you think AGI is just around the corner, try to build an agent.

You quickly learn how dumb these systems currently are. And I mean a real agent, not a Playwright web scraper or PDF summarizer.

But make no mistake, these complex agents will get better. And not just better, they'll blossom into agents that can work for days or weeks unattended without making mistakes.

Whenever I want a more realistic perspective about AI, I read robotics papers or talk to roboticists.  They've been dealing with the messy real world for many decades and they're under no illusions of how fiendishly complex it is to build a machine that can think and act quickly in the real world.

We underestimate just how complex and wondrous we humans are!  We're some of the most amazing biomachines ever created.

But we're already making progress with bots too. Major progress. We've seen a big leap in humanoid robotics capabilities in just the last few years versus the decades before it.  That's often the nature of technological revolutions. They come slowly, inch by inch, and then all the building blocks are there and it comes together very quickly before hitting another wall of diminishing returns. Every exponential curve eventually becomes an S curve. But we're still in the early parts of that dramatic rise in capabilities and we can see it all over the space of robots and agents.

Here we can see O2, from Figure, a California robotics company, doing test work in a BMW factory. 

If you look at this clip from the recent Tesla after party, where the robots are handing out bags and drinks, you can see they already have a strong ability to perceive and interact with the world.  The robots are tele-operated, but that just means that they're collecting lots and lots of data that can then be used to train them on a variety of tasks so they can do those same tasks autonomously.

Here's a Boston Dynamics robot doing work in a factory fully autonomously already. 

Beyond robotics, we're entering an era of Service-as-Software, where more and more boring work gets automated, which will massively speed up what we can do and how fast we can do it.   Doing sales outreach, writing reports, detecting disease in an X-ray, making calls to setup appointments, doing research on complex topics, fact checking and so much more is coming. 

OpenAI has a blueprint for its five steps to AGI:

1. Level 1 ("Conversational"), AI that can converse with people in a human like way

2. Level 2 (“Reasoners”), means systems can solve problems as well as a human with a  doctorate-level education. We've seen the first stages of this with o1.

3. Level 3 (“Agents”) are systems that can spend several days acting on a user’s behalf.

4. Level 4 ("Innovators") are AI system that can develop new innovations and novel ideas.

5. Level 5 (“Organizations”), means an AI systems that can do the work of an entire organization, like a small or medium business

Each one of these steps is a quagmire in and of itself.  Each one has a series of mini-steps, setbacks and problems that are incredibly hard to overcome, not to mention painful integration points to make it all work in any single business or to accomplish a single task.

Right now we're at the dawn of Reasoners, with models like o1-preview and o1-mini.  These will get better and lead to smarter, agentic systems, capable of doing complex work in the digital world and in the physical.

One way these models will get better is that older models will bootstrap new models.  It might take the form of a method pioneered out of China, which uses an LLM as a teacher to a reward driven agent. With this approach, the LLM dynamically sets the reward and leverages its internal world model to help the agent make its initial decisions versus just trying random actions.  That dramatically speeds up the process of building a strong RL augmented agent. 

Even better is that the agent is likely to get stronger than the original LLM.  It rapidly learns from the LLM and then outpaces it because it can then do its own exploration, so it ends up getting a better world model on our task than its teacher.

The student becomes the master.

How we train these systems will change a lot in the coming years.  We'll teach models more through experience and as the model architectures change, they will get better at continual learning and overcome catastrophic forgetting.

One possibility for beating both catastrophic forgetting and enabling continual learning, two of the longest standing challenges in AI, is already here, the Mixture of a Million Experts paper by a lone researcher Xu Owen He, at Google Deepmind.  He built PEER, a learnable router that is decoupled from memory usage.  It's able to scale to millions of experts and some of the routing neurons are only a single MLP (multilayered perceptrons, the digital building block of neural nets inspired by studying human neurons, though radically simplified from their biological counterparts).

From the paper itself:

"Beyond efficient scaling, another reason to have a vast number of experts is lifelong learning, where MoE has emerged as a promising approach (Aljundi et al., 2017; Chen et al., 2023; Yu et al., 2024; Li et al., 2024). For instance, Chen et al. (2023) showed that, by simply adding new experts and regularizing them properly, MoE models can adapt to continuous data streams. Freezing old experts and updating only new ones prevents catastrophic forgetting and maintains plasticity by design. In lifelong learning settings, the data stream can be indefinitely long or never-ending (Mitchell et al., 2018), necessitating an expanding pool of experts."

Essentially massive MoE models like this hold the potential to lower inference memory consumption and beat catastrophic forgetting by freezing or partially freezing weights and just adding in new experts to learn new information. The neural router is much more effective because it's general purpose and learned.  It's like having parallelizable Adaptors/LoRas built right into the model.  It also mirrors the plasticity of the brain in that the router can reconfigure itself as new knowledge is learned, much like your brain prunes or strengthens synapses.

Once you have an ever-learning model like this in place, whether it proves to be this architecture or another, we can move to a practical life long learning system that mirrors how people process massive amounts of information.

Agents also won't be limited to what they learn on their own, like humans.  They can share what they've learned.

Humans can do that too, with language and books and multimedia and YouTube. But those are lossy transfer mechanisms.  I have to explain what is in my head when I'm teaching you and you have to understand it and language doesn't capture everything I've learned or experienced exactly as I experienced it or learned it, even if I'm a fantastic writer.

With AI we can directly transfer memories completely intact between systems with no loss at all.

That's very powerful.

We also might be able to get to robots and agents with downloadable skills by just using Adaptors and hot swapping them in on a per task basis or as needed, which our team at Kentauros AI is experimenting with now on our AgentSea platform and Apple recently used the same approach for their on-device models. 

With hot swappable skills, we can fine tune a small adaptor, like a QLoRa that is specific to that workflow or task. So you might train it up on your own personal in-house application and the training is very fast because you're leveraging the model's general purpose reasoning. Or you might train it to shop on Amazon or do research on LinkedIn.

We don't need a general purpose model that knows everything.  If it can dynamically load adapters on the fly and suddenly it is better at that specific skill, like Neo suddenly knowing Kung Fu.

Think of a cleaning robot that downloads a new adapter on how to do the  dishes or make dinner. Now it can swap in that module when it's time to make that meatball pasta for the weekend.  It can download the dog walking module and learn to walk your dog instantly.

All of that means we will soon(ish) have very strong, very powerful agents.

AI agents will seep into every area of work, online or in the real world, with robots.  AI lawyers, doctors, software engineers, security engineers, dock workers and self-driving business and ships and cars.

They'll remake the very nature of work and how we do it.

Economics and Trade

Economics is one of the areas where AI can deliver a true revolution too.

That's because the dismal science is antiquated and full of theories from the turn of the century, made by old men in smoking jackets using questionable analog statistics. Our economic theories are simplistic, outdated and rarely reflect reality.

The dominant economic theories of today were primarily developed in the 20th century, with roots tracing back to the 18th and 19th centuries. Keynesian economics are still highly influential today and they was developed in the 1930s.  Neoclassical economics comes to us from 1870s and dominated after World War II. At least modernism was developed in the 1970s, with Milton Friedman pushing back against Keynesian economics and emphasizing money supply control but it was still created before the computer and digital revolutions. 

(Source: John Maynard Keyes, United Press International)

While these theories have been updated and refined, they simply can't capture modern economics in any meaningful way.  New economic theories are slow to develop.  Newer theories try to bring in behavioral economics, incorporating psychological insights into economic decision-making, and by viewing the economy as a complex adaptive system, made up of many tiny models.

There is a desperate need for a "MEADE" (Multiple Equilibrium and DiversE) style paradigm shift, which uses a range of models to understand economic phenomena rather than relying on a single unified theoretical framework.

Even that won't be enough.

That's because economics is dense, complicated and largely impenetrable for the average human being.  It's incredibly hard to understand.  We're not wired to easily comprehend systems that move at massive scales with trillions of tiny variables. 

As Richard Dawkins wrote in "The Blind Watchmaker":

"Our brains were designed to understand hunting and gathering, mating and child-rearing: a world of medium-sized objects moving in three dimensions at moderate speeds. We are ill-equipped to comprehend the very small and the very large; things whose duration is measured in picoseconds or gigayears; particles that don’t have position; forces and fields that we cannot see or touch, which we know of only because they affect things that we can see or touch."

• Dawkins, Richard (2015). The Blind Watchmaker: Why the Evidence of Evolution Reveals a Universe without Design . W. W. Norton & Company. Kindle Edition.

And because it's super hard to comprehend, that's exactly why machines will do it better. Machines are better at size and scale, exactly that kinds of thinking that humans are terrible at. 

Over the next decade we'll move to a rapid synthesis of up-to-the-minute economics data and an explosion of new economic theories built from real world understanding as world trade explodes across our screens like winds moving around the world in beautiful dance of delightful chaos and complexity.  We'll see it all as it happens rather than months or years after the fact, when its already too late to make good economic policy to avert the crisis or stimulate a new boom.

We'll watch the flow of money and people and ships around the world in a revolution of real-time economics statistics tracking and we'll see a dramatic rise of reasoning AIs that make better decisions faster than any human being. 

Take something like shipments. The world economy moves by sea, with around 80-90% of the volume of international trade in goods is carried across the oceanic sea lanes. The percentage is even higher for most developing countries. In 2023, sea transportation accounted for 47% of goods traded between the EU and the rest of the world measured by value, and 74% measured by volume. There are over 50,000 merchant ships trading internationally, transporting every kind of cargo. The world fleet is registered in over 150 nations and manned by over a million seafarers.

(Source: Wikimedia Commons)

Tracking all those shipments and goods is a nightmare.  Much of our economic data about it comes from government published reports but the quality of that data varies widely and is often outright manipulated.  Many countries struggle to create high-quality economic data and because of outdated methodologies for measuring GDP and other indicators it doesn't match reality. Developing countries often lack the institutional capacity and funding to conduct comprehensive economic surveys and data collection. 

Surveys are error prone and don't capture even a fraction of the capacity and nuance of the underlying real economics.  Countries will understate inflation to avoid higher interest payments on inflation-linked bonds or make sure international aid keeps flowing.  Greece and Italy almost certainly manipulated their budget deficits to meet Eurozone criteria prior to joining the EU. China regularly inflates economic statistics to always maintain an air of rapid growth.

All of this bad data and lying leads to bad economic analysis and forecasting, which leads to misguided policy decisions.

But we don't need government figures if we can watch all the shipping containers flowing out of ports around the world with machine learning.  That's already happening.  Multiple companies and satellite providers already have software that tracks shipments in real time, watching ships with object detection and image recognition from their eyes in the sky. 

BlackSky and Spire Global have partnered to create a real-time Maritime Custody Service (MCS) that can automatically detect, identify and track over 270,000 vessels worldwide using AI and satellite data.  Mitsubishi is using AI to track "dark fleet" ships from space, violating international laws to fish illegally or move illicit goods.  Satellogic and HappyRobot created a AI models that identify individual shipping containers and different types of vessels at ports using high-resolution satellite imagery.

(Source: Satellogic Blog post: “By selecting an area of interest, such as the port of Hong Kong in this picture, and leveraging Satellogic’s high resolution combined with Happyrobot’s AI models, we were able to monitor port activity over several days.”)

That's just the beginning.

Imagine ambient AI embedded in every level of our economy, watching the flow of money, goods, people, trade, business and immigration, trillions and trillions of data points flowing into massive economic models and massive reasoning engines.

As more and more money becomes digital, it becomes easier and easier to spot wide scale spending patterns, economic data, and to predict economic crashes and booms long before they ever happen.  In a few decades we're likely to see AI recommending stimulus and lending rates and minimum wages based on real-time, adjusting the economic levers to smooth out the downsides and supercharge the upside. 

The biggest limiting factor will be how much we resist the switch to the real-time information economy.  How much will we fight and resist letting our brilliant machines track the ebb and flow of money and people?  How much will we battle back against better decision making due to misguided fear and a desire to keep our thumbs on the scale out of blind animal panic and misguided delusions of "losing control" of our AI. 

Politics and fear are that only things that can slow down the inevitable move to the age of "algorithmic economics."

It might take many decades to see the age of info economics, because resistance in government is massive to this kind of transparency.  There's power in manipulated statistics. But we'll likely see nations monitoring other nations first, rather than themselves, and that will force transparency whether nation-states like it or not, as every nation monitors every other nation and eventually forces direct transparency.

The idea of algorithmic economics scares some people but we should welcome it because the machines will make better sense of the world economy than we ever will. It's simply too complicated for any one person or group of people to understand but its exactly the kind of problem that machines tackle better than any bio-mind.

We could look at a lot more industries and world changes. I could write an entire novel on this and probably will. Every industry from chip making, to knowledge work, to software creation, will shift massively in the coming years. 

But now that we've looked at the positive lets look at the downsides, in a realistic way, before wrapping up.

Every light has a shadow.

If we want a balanced view of the future we can't just look at the bright side.

We have to stare into the abyss as well.

(The Horror. The Horror.)

He Took a Face from the Ancient Gallery

While the AI doom cabal scream about drug dream fantasies of AI going foom or taking over, or us "losing control" to magical mega-AIs, something far wickeder is festering right under our noses in the real world.

AI can and will supercharge war and surveillance to terrifying new levels. 

To see the most horrific use cases of AI you have to turn away from the private sphere and look at governments, especially authoritarian governments, but also just well meaning democratic ones too.  Nation states have enemies and their basic nature is to protect their citizens and that often leads to ends-justify-the-means kind of thinking.

The AI doom movement has concentrated their fire almost exclusively on strangling and limiting the development of civilian AI while giving an absolute and total free pass to government and military uses of AI. This is a looming disaster if they win because it means civilians will have no means to fight back against all powerful states that can peer into every aspect of their lives.  As we get more and more digital it becomes harder and harder for any person to move through life without leaving a long digital trail in the clouds of machines we interact with every day and that means nobody will escape the all seeing eyes of the powers that be.

As we move to nation-state digital currencies it will be easy to cut someone off from their money like flipping off a light switch. AI driven camera networks will watch every aspect of people's lives and they already do in China.  Chinese surveillance systems can even ID people wearing a mask.

To be fair, we've seen a few scattered attempts at bringing awareness by activist groups, like the excellent Slaughterbots sci-fi short film that predicted the rise of the AI driven drone explosives that have sadly become real life in a vicious and drawn out Russia/Ukraine war that echos the cyberpunk fiction of the 1980s. 

(Source: Reuters special report on drone warfare in the Ukraine/Russia war)

But overall these efforts are few and far between and these groups have gotten zero traction on any bills that will restrict the ever accelerating AI arms race of governments all over the world for new and more powerful weapons and surveillance systems. 

There's a reason they've gotten zero traction.

AI is just too tempting for governments and there is zero chance that any government will willingly restrict its own growth and risk losing the edge to rival nations as this cartoon from Economist shows so brilliantly:

(Source: Economist KAL cartoon after the UK AI Safety Summit)

Every bill coming out of America, China and the EU (including the EU AI Act which is starting to go into effect) is strictly focused on civilian AI and gives a 100% free pass to governments to do whatever they want.

Even if any government does sign a bill to restrict AI in weapons tech or surveillance, they'll just build it anyway with black budgets as we've seen time and again with projects like Total Information Awareness (TIA), which was explicitly defunded by congress but a decade later led to the very mass surveillance tools that Snowden leaked.

To refresh your memory on TIA, the scope of those systems included "a 'virtual, centralized, grand database',^[38]  the scope of surveillance included credit card purchases, magazine  subscriptions, web browsing histories, phone records, academic grades,  bank deposits, gambling histories, passport applications, airline and  railway tickets, driver's licenses, gun licenses, toll records, judicial  records, and divorce records,^[8][12] along with "health and biological information TIA collected included drug prescriptions,^[8] medical records,^[39] fingerprints, gait, face and iris data,^[12] and DNA.^[40]"

Now imagine what they can do with AI embedded into those systems?  Imagine what a sadistic regime like the regimes of Mao, Satlin, Hitler, Mussolini and Franco can do with that kind of dark intelligence monitoring every aspect of their citizens lives. 

"If you want a picture of the future, imagine a boot stamping on a human face— forever," wrote George Orwell in 1984.

And nothing will make that shadow more clear than the dawn of the AI arms race.

People have started to misuse that term to mean companies racing each other to build better products, like OpenAI and Anthropic. That's absurd nonsense. When I say arms race, I mean exactly that, "arms" as in weapons of war.

The next five to fifteen years will mark the beginning of a new digital Darwinism.  It’s survival of the fittest at the military and nation state level.

Today’s narrow AI feeds on big data. The more data you can collect the smarter your AI gets, at least until it hits a point of diminishing returns after it's chewed through petabytes of imagery and text and video.  If your citizens don’t have any right to privacy and the government can do whatever it wants, it makes it super easy to build massive data sets to train your AI.

That gives authoritarian regimes a tremendous early advantage.

While the EU is still debating AI ethics, holding hands and singing songs, and passing laws to tie up their own companies in red tape, China started rolling out AI at a breathtaking speed five years ago. As early as 2019, Alibaba-built “City Brain’s” that could already monitors 50,000 cameras, directs traffic, detects car crashes in under 20 seconds and tracks criminals.

They won’t stop there.

(Source: Alibaba website: Alibaba City Brain Dashboard circa 2019.)

They want to build city scale visual search engines, call fire trucks before the people even know their house is burning and they most assuredly have stealth purposes not listed in their marketing material, like tracking dissidents and other enemies of the state.

That’s the biggest problem with AI.  No matter what we can do it can always been used for good or evil.

If we use it to track old people in hospitals when they go into cardiac arrest or stop breathing, it’s only a short hop to monitoring old folks for health insurance to penalize them for not working out enough or taking their medicine.

For governments the ethics implications are even worse. Every country will want to deploy AI at the City Brain scale but they’ll need to tread a fine line between privacy and the public good.

That line is fuzzy at best.

The road to Hell is paved with the motto “for the greater good.”

For every use case I can think of that’s a clear public win, I can think of one with a more sinister edge.

Yes, there are clear and obvious greater goods:

• Spotting fires faster than any human ever could means firetrucks get their faster.

• Detecting violence and catching violent criminals means less people die from grisly gunshot wounds or a knife in the back.

• Recovering stolen cars and purses means a lot cheaper insurance.

• Directing traffic means a lot less time commuting.

Sill, a system that spots criminals and “bad guys” can easily spot dissidents and political enemies and anyone who stands up and speaks out against the system. Expect governments to use the now standard bogeyman of terrorism as a justification to sneak these technologies in under our nose without a public debate. They won’t even have to sneak them in because they’ll already be there directing the morning commute and calling ambulances so politicians with darker agendas can easily layer in a second purpose.

(Source: John Jonik)

And as cities get smarter, it’s not hard to see non-lethal and lethal autonomous arrest technology. A drone will swoop down and capture the bad guy with a net or a glue spray or just kill him with good old fashioned bullets.

The only real question is, who gets to define the bad guy?

There are obvious bad guys like murderers. But it doesn’t take long to start seeing everyone on the other side of the political debate as bad guys too.

Who the bad guys are shifts with who’s in power.

If you have a just government, making just decisions, then the bad guys are people we tend to agree on. But an unjust government can mark anyone who disagrees with them as a criminal.

And that’s the biggest problem. The real bad guys, authoritarians, the vicious and the cruel will already have a dual use technology ready and waiting to serve them when they come to power. AI will live in our public cameras to get help to people faster but its latent power will be there waiting to be activated.

Today’s AI serves its masters faithfully. Unlike a soldier, it won’t ever question the order of its commanders. It will just do what its told with ruthless efficiency whether its right or wrong.

Many fear AI with consciousness.

Maybe we should fear the exact opposite?

AI without consciousness.

I don't fear AI taking over. I fear dumb, narrow AI, controlled by nasty, sadistic, power hungry people. AI is just a tool.  And tools in the hands of bad people are the real problem.

Humans are the real monsters.

Balance is Best in All Things

We’ve looked at AI detecting cancer, running entire cities, creating astonishing new film and TV, creating software that can do any kind of work, making smart weapons smarter and more lethal, and robots that can do the dishes and fold your clothes, robo labs and scientific breakthroughs, and a ton more.

While some people wonder what happens when we automate more and more work and AI seeps into every aspect of our life, the answer is the same answer it's always been.

We'll adapt.

It's what we do.  We evolve. We change.  The people alive today are not the same as the people alive just a generation ago and we live a light years different life to people 15 generations ago and radically different from the people 100s or 1000s of generations ago.  As Tim Urban wrote in his brilliant essay, Meet Your Ancestors, All of Them, your great grandfather to the 20th power would look at your life now with shock and distain. 

"Your great²⁰ grandfather kept it real. When he wasn’t torturing somebody, he was being tortured himself. When he wasn’t catching the Black Plague and dying, he was slaughtering women and children in the Crusades. And weirdly, he might have had the same last name as you.  If he could meet you, he’d be blown away by the ease of your current pussy existence. But not as blown away as your great⁵⁰⁰ grandfather would be."

(Source: Tim Urban essay on your ancestors, linked above)

We've adapted to technology change millions of time in our history and we'll do it again. Human life is not just one thing.  "Always in motion is the future," said Yoda and always are we too.  We're not running around in the forest collecting things and living in communes like we did in the hunter gatherer era. We're not hunting the Water Buffalo.  I'm sitting in my bed writing this essay on a computer, a near magical marvel of the modern world, the work of millions of my ancestors and modern businesses.

Nobody is clamoring for a return to hunting and killing whales so we can dig the white gunk out of their heads to make candles.  We use electricity now.  The whale hunter jobs are gone but we adapted.  And it's better that we can just flick a switch and light the world.

In many ways the paranoia and ridiculous fear mongering around AI mirrors early fears around electricity.  It's dangerous and terrifying and it will kill everyone if it gets more widespread! And it's not that there were no problems, like poorly insulated wires and lack of standards. There were. Real people did get electrocuted before we got very good at stringing wires around the world. Now the dangers are vanishingly small unless you do something magnificently stupid like putting a fork into a light socket.

AI will change the world. It just won't happen as fast and as suddenly as many people fear.

Think of this essay as playing out over many decades, not next year or next week.

While people are predicting AGI is just around the corner, anyone who works with AI agents like my team does knows just how stupid these models really are and how limited they are in the real world.  We live in a weird world where AI is smarter than many PhDs and yet dumber than your cat.  They can tell you about nuclear physics and speak 100 languages fluently and yet they lack any and all common sense and they can't seem to get tasks done without going completely off the rails.

This confuses a lot of people. If it can speak a 100 languages it must be smart. But it's just dead wrong. It's only part of the story. These models are smart and getting smarter but they're still dumb too.

Here's an example from Max Bennett's amazing book A Brief History of Intelligence.  I started with the line "He threw a baseball 100 feet above my head, I reached up to catch it, jumped..." and let it finish the sentence.

Because of your common sense, you'd know the ball was too high to catch.

GPT-4o does not know that (unless they fine tune it on that now, which they might do.)

Don't get me wrong. Many of these systems are still good and useful and they do things that simply weren't possible a few years ago and they're only the beginning of what's to come. I absolutely love coding with AI. I'm a pretty terrible programmer but I'm a great idea person and now I can prototype ideas without waiting for engineers to tell me "no" and "it can't be done."

It can. And my lack of experience is an asset in this case because I will try anything and AI lets me experiment quickly with ideas, discarding the ones that don't work and digging deeper on the ones that do.

AI is here. It's evolving but it won't go foom or change the world an instant. It will change it and we'll adapt and for the most part it will be a good thing.

More intelligence is always better.

Is anyone sitting around wishing their supply lines were dumber or that it was harder to detect diseases? 

Do we really want dumb cars? Autonomous cars will save millions of lives each year. 

Turns out humans are terrible drivers.

It takes perfect concentration to drive a car right. And that’s something humans just don’t do with their ever-deteriorating attention spans.  Globally in 2019 1.25 million people died in car wrecks. That’s 3,287 deaths a day. More than 20–50 million more are injured or disabled.

Still humans are under the delusion that, like Rain Main, they’re very, very good drivers. Humans have the great ability to fool themselves that way.

In the early days of autonomous cars, every fiery self-driving car crash made the news.

In a short time, no self-driving car crashes will make the news because they'll just be much better at it than humans who killed so many more people with their inability to pay perfect attention to ever changing road conditions.

By the 2030s the death toll will plummet. Steering wheels, once required by law, will disappear all together.

(Source: Mercedes F 0150 concept vehicle.)

The inside of cars will be redesigned as mini-apartments on wheels, with tables and chairs facing each other.

Drinking and driving will become a good thing.

I once tried to describe what I did to my 86 year old grandmother. When I told her AI, she said "what's that?" I tried to explain it and then showed her a self-driving car. She said "I'd never get in that."

And I told her, "Grandma, tomorrow's kids will say the exact opposite, I will never get in the car with a human behind the wheel because it is just too dangerous."

Change is coming and its coming fast.  Just not as fast as everyone thinks.

As I wrote earlier, we're still in the early phases of level 2, reasoners, with models like o1. They will get better and better over time. But it will be more gradual and slower. In some areas it will rocket ahead and in other areas it will be slower and steadier. Technology diffuses into the world at differing speeds, like water rushing fast at first and then trickling slower as it spreads out and runs into obstacles and barriers.

In other words, there is friction at every one of these steps.  A lot of it.

We tend to overestimate the speed of some things and underestimate it in other ways. There’s a great line in Black Hawk Down, where a soldier asks how long it will take to fix a helicopter?

“Five minutes.”

“Nothing takes five minutes.”

But does it really matter? Five years or fifteen or fifty is a drop in the bucket for radical changes that will remake the world.

Change is coming and it’s coming relatively fast and it will only accelerate.

For first time in history we have a technology that can do just about anything. Just like us.

We're about to see a Cambrian explosion of intelligence.

And life will never be the same.

If you think AGI is just around the corner, try building an agent.

And I don't mean a "talk to your PDF" or scrape this website with Playwright agent.  Those are good, useful tools, but they hardly qualify as complex thinking and reasoning systems.  I mean try to build a real one. 

What's a real agent?

An AI system that's capable of doing complex, open ended tasks in the real world.

Even more complicated?

Build an agent that can do long-running, open ended tasks.  Like a person.

The real world could be the physical world or online.  What we're talking about here is when agents get out of the lab and face the never ending complexity of life.

Life is messy.

Try to build an agent that deals with the barely restrained chaos of real life and you'll realize just how dumb these systems currently are.

If you work with agents every day, like my team does, you quickly see the limitations of today's frontier models.  We see the same problems again and again and we divide those problem into three broad areas that we affectionately call:

• Big brain

• Little brain

• Tool brain

"Big brain" means anything that deals with higher level reasoning, strategic long and short term planning, abstraction, expert understanding and common sense.  As we'll soon see these are the biggest problems we run into and we'll talk about some potential fixes but this is the hardest area to make stronger.

Here's a simple example that seems trivial to you but is hard for machines: solving the problem of "I'm thirsty."

You run through scenarios like I could have a Coke, or OJ, or water. Do we have enough filtered water in the house?  No, I have to run to the store. Do I really want to run to the store just so I don't have to drink tap water? Yes. Okay, I have to put on my jacket, take my wallet and keys, go to the store to buy filtered water so I can satisfy my thirst. 

There's a cascade of steps here from making a decision, getting dressed, taking your keys and money, closing the door, locking up, heading to the store, finding the water, buying it and drinking it.

Moravec's paradox says that reasoning for robots is relatively easy and sensorimotor perception is hard.  That's wrong. Both are hard, especially when you try to generalize reasoning.

We can also think of this as strategic reasoning.

"Little brain" means the thinking and planning of precise in-the-moment actions or steps, like a UI navigating agent that needs to find the right button or a robot that has to decide whether the thing in front of it is a tree or a stop sign or a half broken wall. 

Each step on our journey is a cluster of sub-actions specific to the major milestone we're on. If I'm checking out with my bottled water, there's picking the shortest line, waiting on that line, stepping forwards, answering questions like if I want a bag or not, small talk with the checkout person, and how do I want to pay? 

You can think of these as an atomic sub-cluster of reasoning tasks. Another way to think of it is tactical reasoning.

If I want to drink my drink, that's another cluster of actions that "Little Brain" handles.  At the unconscious level my brain says I need to reach out my arm, open my hand, close it around the bottle firmly, pick it up, twist off the cap with my other hand, tilt the bottle back to my lips.

"Tool brain" is the quality and precision of the tooling/appendages/sensors/APIs/real world interfaces given to the model/robot to execute an action. It doesn't matter if the little brain figures out it needs to click the "search" button if the tool itself to return those coordinates is problematic.  It's like having a broken hand.

You may know you want to pick up the bottle in front of you, as well as the right steps to do it but if your hand isn't working right you can't pick it up.

None of these problems are easy to solve. None of them work "right out of the box" with today's foundation models. Some of them are made easier by the emergent capabilities of these models but there's a tendency to believe these models are magic that can do anything easily with no work involved.

It doesn't take very long working with agents in the real world to be disabused of that notion fast.

We're going to talk about each of these and we're going to talk about how we can fix them in the future and what it will take to do it.

In essence, every agent today is a beautiful snowflake, a complex mash-up of multiple models, human integrated logic, heuristics, fine tuning, error correction and scoping, not to mention chewing gum, glue code and fantastical, multi-layered prompt engineering.

Let's take a look at big brain reasoning first and why it's the hardest part to solve.

Big Sky Mind

(Image: Midjourney 6.1)

When we're talking about reasoning we're talking about the "thinking" capabilities of today's foundation/frontier models like o1-preview/mini/o1, GPT-4o, Claude 3.5, Llama 3.1 405B and Gemini.  Usually we're calling these models through an API to a proprietary cloud service, though we might be running LLama 3.2 in our own private cloud or datacenter.

There's a cost with the round trip time to the cloud and a cost imposed by how brilliantly or how crappy these cutting edge models make decisions.

So how do they make decisions?

Amazingly and then horribly.  In an endless loop. 

Even worse, as the task gets longer and more involved their errors cascade.  It's a compounding problem.  A mistake in one part of the task can carry down into other parts of the task and corrupt all the steps after.

A classic example is that the top models today tend to hallucinate the old version 1 style of OpenAI's API to talk to the model. The old style was written like this:

response = openai.ChatCompletion.create(

The new way is written like this:

response = client.chat.completions.create(

These models will routinely swap out the old method for the new and now every step they took that is related to that mistake is wrong. 

This is just a simple example but one that happens all the time, even when you give the model the latest documentation.  It will often switch to the new style after you correct it, but after 10 or 20 steps its contextual memory buffer will overflow and it will revert back to the old way of doing things. 

This happens even with reasoning models like o1-preview and o1-mini.

It often feels like working with a brilliant idiot.  These models are capable of incredible decision making at times and then they just do something that makes absolutely no sense.

In this experimental version of our next-gen GUI navigating agent, you can see it make generally good decisions and some dumb ones too.  The prompt was "Go to Wikipedia and tell me the countries that Edward the Confessor was king of during his lifetime."

In the video the browser was already open because I was running multiple variations of the test. The agent correctly reasoned to use Ctrl+L to automatically highlight the URL bar. It then correctly guessed the page, either pulling from its memory system (which I'm testing here) or just by deducing the general structure of the page names.  All good.

Then it decides it needs to click the "language" setting for some reason and to scroll up and down for reasons unknown, instead of just reading the page. 

Even worse, it hallucinates several "tools" that it does not have access to, despite having a list of tools to choose from that help it find GUI elements, decide coordinates to click on, take screenshots and more.  It decides it needs to use the "read_text" tool which does not exist. It straight made it up.  The main model running here is GPT-4o and it forgets that it has built in OCR and reading capabilities and that it doesn't need an external tool to read anything.

At one point I just hacked the hallucination by creating a dummy tool called, read_text that returned text that said "You DO NOT need this tool. Don't use it. You have a built-in ability to read text as a powerful frontier model. Just read it. Don't call this again."

That made it worse.

The model would often just get stuck in a loop calling read_text or analyze_page or any other set of fake tools it could imagine in an endless loop, failing to complete the task, even though it already had the answer (from reading it!) and all it had to do was trigger the "I think the task is done" loop to give the answer to the user and end the run.

In other runs, I watched the bot fail to open the web browser, something it successfully did 99.5% of the time.  It decided that the "web browser" was the "agentsea home folder" for reasons that only its digital neurons know the mystery of and it failed to get past stage one of a task it had completed a 1000 times before with ease.

Welcome to the absurd reasoning of the big brain. 

Every team I've talked to that's building agents has pivoted from "we'll build a general purpose agent" to building a "narrow agent" where they've drastically pruned the scope of their ambitions and surrounded a Multimodal Large Language Model (MLLM) with expert heuristics and guardrails and smaller models to compensate for the absurd reasoning of MLLMs.

What do I mean by absurd reasoning? 

I mean these machines completely lack basic common sense, not to mention consistent and clear deductive and inductive reasoning and they struggle with strategic thinking and building a complete world model.

Common sense and lack of a world model are perhaps the biggest problems. 

"Common sense" is a series of built-in abstract patterns that we recognize in almost any variation or configuration without even realizing it.

A "world model" is an abstracted understanding of the world around us. If you get cut by a knife one time, you abstract the concept of "pain" and "sharp."  Now you can see any jagged object like a rock, a spike on a fence, a pike, a sword and you know it’s dangerous.

A multimodal model has to see pictures of a knife from the side, low angles, high angles, overhead, front and back during its training or it has no idea this is the same concept and should be grouped together and it can't generalize what it learned to other sharp objects easily.

Meanwhile, a fish knows this automatically (and so do you).  You can show a fish a picture of a frog and train it to get food if it touches that picture.  Now give the fish a new picture of the frog at a different angle and it will instantly know that it's still a frog and swim right up to it to get that food.

An experiment in 2022 by comparative psychologist Caroline DeLong at Rochester Institute of Technology did the above experiment with goldfish.

As Max Bennett writes in his amazing book A Brief History of Intelligence: "Amazingly, the fish swam right up to the new frog picture, clearly able to immediately recognize the frog despite the new angle."

In other words, the fish (and you) have a built in understanding of 3D that it can generalize to any object it sees from only a single angle.

So how does this lack of common sense show up in agents?

Let's say the agent misclicks a form.  It meant to click the name field and misclicked. 

What does it do then?

Sometimes it figures out it didn't click the form but other times it just starts typing anyway!  Even worse, sometimes it doesn't seem to recognize that it failed to fill out the form, even from an updated verification screenshot showing that the form field is empty.

Here's an agent about to click the submit button on Wikipedia even though it never managed to type anything in the search field.

A human just would almost never do that.

You would know the field was empty and something went wrong. (Well most would. You should see some of my old IT consulting clients but that's a story for another day...)

Besides my ex-clients I say almost because sometimes someone has no software experience whatsoever or they never learned about interfaces.  But most people who've used a search engine or Wikipedia before would never make this mistake except under extreme circumstances like they're badly drunk or trying to do too many things at once and not really paying attention.

We take common sense for granted but it's a powerful map of the world and a quick and easy way for our brains to understand what's happening without having to burn too much energy.  Common sense delivers lightning fast micro-understandings that happen just under the surface of our consciousness, helping us every single second of every day. 

Your brain is constantly monitoring for tiny micro-signals that something is "right" or "wrong" and it happens seamlessly, easily and automatically.

If you were naked and I told you to run down to the store you'd have the common sense to know you should probably put some clothes on, even though I never explicitly told you that.

Here's another example from Max Bennett.  I started with the line "He threw a baseball 100 feet above my head, I reached up to catch it, jumped..." and let it finish the sentence.

Because of your common sense, you'd know the ball was too high to catch.

GPT-4o does not know that.

Usually.

Every once in a while it gets it but that's the nature of these systems. When it comes to task consistency, every once in a while doesn't cut it.

Don't get me wrong. Many of these systems are still good and useful and they do things that simply weren't possible a few years ago.

I wrote one knowledge assistant that went out and crawled 30K websites of potential partners and the bot summarized what they do, found their current pricing and scored them according to a hand crafted selection criteria in a function call.  It saved our team a massive amount of time. It would have taken us months to go read all those sites manually. 

But the bot also sucked too.  It was like playing edge case whack-a-mole. It was not easy to get right and it took about a month of hacking away and adding checks and balances and prompt engineering to make it work.

That's still a far cry from just talking to a super smart machine and giving it the broad strokes of what you want and having it go out and figure out the whole task on its own and do it effectively with minimal supervision, checking in every few hours or days with questions, follow ups and adjustments.

This is not "Her."

Every single agent today is a bespoke work of art, a beautiful little snowflake that some team slaved over for months to get working.  They're hand crafted marvels of overlapping traditional AI methods, prompt engineering, creative thinking, hand written code, hacks, heuristics, and chewing gum to hold it all together.

And no GPT-5 or Claude 4 are not just going to "solve everything."

They won't.

Neither is GPT 6 or Ilya Sutskever's SSI and its "straight shot to superintelligence."

And now we know, neither will Project Strawberry, aka o1, o1-preview and o1-mini, just as I predicted in my article Why LLMs are Much Smarter Than You and Much Dumber Than Your Cat a few months ago.  They're awesome models but they're only good at learning the hard reasoning of math and science and are not good at the fuzzier reasoning of people.

But it's not just higher level reasoning where today's agents fall short. 

And that's where Little Brain comes into the picture.

Little Brain, Big Problems

Reasoning about sequences of steps is partly a Big Brain problem and partly a Little Brain problem. 

Little Brain reasoning problems can be simple or complex. Simple is just knowing the correct action to take.  Complex is not knowing the right chain of actions to take.  In one test of our GUI nav agent, I had the model open Gmail, write a poem and send it to the team.  It did all that.

Then it made a Little Brain tactical understanding mistake.  It decided that it needed to verify that the email was sent, because we hadn't given it a memory of previous actions yet, so it had no memory of what it just did. 

It decided that in order to check that the email was actually sent it needed to click on "drafts" instead of the "sent" folder. 

That's wrong reasoning about what to do, so the mistakes can easily compound from there.

In this case it was hilariously wrong.  It noticed that there were 3 previous failed attempts in drafts from earlier tests and it proceeded to start sending all of them, in a perfect example of terrible Little Brain reasoning.

Common sense also comes in at this level too. 

When you're clicking around a GUI, you have an intuitive sense of whether it's giving you the right "feedback" after you did something.  Your brain is always running a background heuristic to tell you if what is happening is "right."  You don't know where that sense comes from, but you just know.  What's really happened is that your brain has generalized the patterns of "using software" and "sending email" and you are drawing from that rich knowledge in parallel as you take actions.

If you click on a button, but it didn't light up or get highlighted in some special way, you know it didn't get clicked because the outcome didn't match your expectations, even if you're not consciously aware of it. 

If you tried to type in the Departure field on Google Flights and no numbers show up, you know you didn't click on the right part of the form.  If you start typing and the first letter doesn't show up, you go back and correct it before continuing.

Frontier models often can't do any of this very well. They have no sense of inherent rightness of the feedback from their tasks.  Some of that is a tool problem, which we'll get into in a moment, but some of it is just Little Brain reasoning. 

Another example of broken tactical reasoning came to me as I was watching my GUI nav agent try to navigate a calendar on Google Flights.  At one point it tried to click on the "departure" date field and misclicked and ended up on "return."  Instead of going back to "departure" it decided that it would select the "return" date number since it was already there and then it would just click the "departure" date right after.  But the interface was designed for humans and human reasoning.  Clicking the return first and then the departure date is just something we never think to do. We go in order of departure and then return.  This habit and reasoning is so engrained in humans that the designers of the interface never even considered someone would click "return" first.

So what happens?

As soon as the agent goes back and clicks the departure date, it resets the return date and all that work was wasted because the workflow just never anticipates that anyone would ever do that. Now it has to click the return date again.

Here's one I saw today:  The agent misclicked on Google Flights and typed the departure destination into the departure date field.  Many calendars don't let you type but the Google Flight calendar does.

That was partly a problem of Tool Brain because the click model returned the wrong field and it happily clicked that field.  But the real deeper problem came with Little Brain.  A mistake like this is recoverable but you have to realize it's a mistake. 

A human would figure it out instantly. 

But the agent? 

It got a verification screenshot in order to make sure what it just did was right.  Here's what it thought when it saw that screenshot:

"The 'From' field has been successfully updated to 'JFK,' indicating that the subtask to enter 'JFK' has been completed. Additionally, the calendar input suggests the next step of entering the corresponding date."

It's not always this bad.  Sometimes these model know exactly what to do. Their built in reasoning is good enough and they know the general steps to take to navigate a GUI to do a basic task like look up some information on Wikipedia.

But sometimes they still can't do it because their hands are broken or they're blind.

The Right Stuff

We take for granted the magnificent tools that evolution gifted us. 

Our eyes are evolutionary masterpieces, capable of fast fine grained tracking of objects and processing massive amounts of information and knowing exactly what to focus on. 

"Your eye can detect over 10 million color hues" and process "36,000 pieces of information per hour" and "24 million images in your lifetime."

And what of your ears, your perfect ears?

They're masterful, in-hardware audio transmutation masters.  They're Fast Fourier Transform marvels, able to convert sound in an instant and send it to our brain for instant understanding. 

I could talk endlessly about the beauty of hands and fingers, the most exquisite fine grained object manipulators in the known Universe. Our hands are a single, universal tool that lets a baseball outfielder snatch a ball out of the clear blue sky, whirl on a dime and hurl the ball across 400 feet of glittering green grass and hit a catcher right in the glove at home plate with unbelievable precision. 

As the ball leaves the player’s finger tips, even a tiny fraction of an inch to the left or the right will change the arc of the parabola of the ball and send it wide right or wide left and yet professional baseball players can make this throw with almost superhuman ease, hitting the catcher with tremendous precision.

That same tool let Da Vinci paint the Mona Lisa, his fingers delicately weaving the soft lines of her smile.  We use the same tool to eat, take out the trash, fold clothes, swing a sword, fire a gun, do pushups, write beautiful poetry, swing a hammer, type on a keyboard and so much more.

(Image: Midjourney 6.1)

With tools like that, is it any wonder we can do wonders?

But our agent tools are not so refined just yet.

This is no surprise.  We haven't been at it all that long and evolution has been at it for billions of years. 

Roboticists have been discovering the pain of building precise manipulation appendages for many decades.  They have a much healthier understanding of the scope of the problems that software only agent developers are just starting to figure out now.

That's where Morvec's paradox was coined.  Basically all the stuff that happens at the unconscious level, that we take for granted, is incredibly hard to model in AI systems.  As we saw earlier, the paradox is only half right, because it is all hard.

Actually, I'd argue we've made more progress on sensorimotor perception and locomotion than we have on reasoning over the past forty years but precision tooling is still a major problem in physical agents and online agents too.

In my article on our second gen agent, Robbie G2, I talked about some of Robbie's tools.  We've refined them since then and given him more tools to use but the tools still make mistakes.

One of our tools we call "grid" because we layer a grid of numbers over an image and ask the foundation model what number is the thing you're looking for closest to?  Then we can zoom in on that area and ask the question again, zeroing in on what the agent is looking for on that step.

It works because while today's foundation models aren't particularly good at giving precise coordinates they're good at giving approximate locations, like "the compose button on twitter is near the number '7'."

It generally works pretty well overall but we noticed Robbie G3 was still struggling with calendars.  After looking deeper at the problem, we saw that it was multilayered but the biggest problem was a Tool Brain problem.

On the third and final zoom in, the red circles were blocking or partially obscuring many of the numbers, so the agent really had no chance of picking the right number.

We ended up crafting an entire calendar doctor sub-tool that can detect a calendar with an OCR model and reconstruct it mathematically, even when it doesn't properly OCR every number.

Better tool. Better results.

Tool Brain problems can be subtle and easy to miss too. 

One of the biggest Tool Brain problems right now is that Robbie is always looking at the past.  That's because our agent is not really looking at the screen in real time like we do.  It's looking at a screenshot, which is a snapshot of the past, like looking at the light of a star. 

When you look at a star, you're not really looking at the star as it is now, you're seeing it as it was millions of years ago, because it took that long for the light to travel to you.  Screenshots are not a million years old but they are the past.  Sometimes something changes before the agent can see it in the time it took to generate that screenshot, pass it to the foundation model in the cloud, reason about the next action and take that action.

A sudden pop-up may obscure a view that was visible, but the pop up showed up just after the screenshot was taken.  Multiple screenshots help to give the agent a sense of transition but it's still not real time.

The agent is not watching itself type in the way that you are and until we have agents that can process video in real time we won't be able to build that kind of feedback into the agent. 

Tesla had to move most of their object detection and object recognition of trees/cars/roads/signs/people into hardware to get the best real time view of the world, processing video at 30 frames per second. That's the very definition of better tooling.

Our tools will get better over time.  So will Big Brain and Little Brain. 

But what does it take to get there?

Dancing Strawberries and the The Bitter Lesson Strikes Back, Again

Project Strawberry, aka OpenAI's o1 models, points to a partial answer and so does the bitter lesson.

A complete answer will take new algorithms and breakthroughs in machine learning.

But that doesn't mean we can't get much better agents today.

The three keys to making progress are:

• Reinforcement learning (RL)

• Learnable/generalizable algorithms

• Scale

RL may not give us magical AGI (to the great disappointment of many AI hype masters who thought Strawberry might be some mystical, supernatural AI from the future) but it can give us better task management, better task understanding and better agents.

Strawberry is a reinforcement learning technique where OpenAI trained the model on Chain of Thoughts, aka "let's think about this step by step." It helps the model correct errors, see its thoughts and critique them and correct them much better than earlier models.

As for the Bitter Lesson of Richard Sutton: 

"The biggest lesson that can be read from 70 years of AI research is that general methods that leverage computation are ultimately the most effective, and by a large margin."

Most folks think that Sutton just meant scale is king, aka GPUs go burr, but it's a bad misreading of what he said.  Scale is important, both on training and inference. But there's much more at work here. Later in the same post he writes a strong clarification:

"AI researchers have often tried to build knowledge into their agents, 2) this always helps in the  short term, and is personally satisfying to the researcher, but 3) in the long run it plateaus and even inhibits further progress, and 4) breakthrough progress eventually arrives by an opposing approach based on scaling computation by search and learning."

The example of the power of pure learning and search is illustrated perfectly by the evolution of AlphaGo to AlphaGo Zero. 

Most folks know the story of AlphaGo and how it beat Lee Sedol, the best player in the world 4-1 but they don't know the story of AlphaGo Zero.

AlphaGo was a supervised learning and exploration system. It learned policies from human games and it had a lot of expert logic tuned into it over time.  It was an incredible system.  But what came after it was more incredible.

AlphaGo Zero had only the basic rules of Go to start from and then it played itself over time, spawning new versions of itself and evolving its understanding through experience and trial and error. It had no human games to learn from or advanced expert characteristics lovingly woven in by its creators.

After only 3 days of training, AlphaGo Zero smashed AlphaGo in head to head competition 100-0.

(Image courtesy of the AlphaGo Zero blog)

The team at DeepMind describes it well:

"[AlphaGo Zero] is able to do this by using a novel form of reinforcement learning, in which AlphaGo Zero becomes its own teacher. The system starts off with a neural network that knows nothing about the game of Go. It then plays games against itself, by combining this neural network with a powerful search algorithm. As it plays, the neural network is tuned and updated to predict moves, as well as the eventual winner of the  games.

"This updated neural network is then recombined with  the search algorithm to create a new, stronger version of AlphaGo Zero,  and the process begins again. In each iteration, the performance of the  system improves by a small amount, and the quality of the self-play  games increases, leading  to more and more accurate neural networks and  ever stronger versions of AlphaGo Zero.

"This technique is more powerful than previous versions of AlphaGo because it is no longer   constrained by the limits of human knowledge. Instead, it is able to learn tabula rasa from the strongest player in the world: AlphaGo itself."

For Big Brain agentic systems, this means training the model to understand tasks and actions with reinforcement learning and that's basically what OpenAI did with Project Strawberry.

Strawberry points the way to what every lab and applied AI team will be working on over the next few years.

Again, Strawberry is a great idea, but it's not some crazy new AGI technique.  The techniques behind the model comes from an old Google Deepmind technique that taught models to play Atari games with reinforcement learning in 2013.

If we look at the Q* paper, it's basically a learned deterministic policy, which means that exploration drops to zero after a time. That's good for games, as it means that the model will never just randomly go right when it's found the optimal path at a certain point is left.  It also makes it good at hard reasoning, like math, or formal reasoning like logic.  You want the model to learn 2 x 2 = 4 every single time, not some of the time.

What this technique can't do is capture "fuzzy reasoning," the kind of "rule of thumb" reasoning that we do. You might make an unconscious rule in your head that to check that an email was sent you need to look in the "sent" directory, but not always. On some systems, ironically our agent that checked drafts was right because of someone's poor UI design.  Sometimes it might be "drafts" or something else and you adapt as needed.  A good generalizing policy needs to take into account the idea of "folders" where emails that have already gone out are stored. 

At Kentauros AI, we see a trajectory to advance our own autonomous GUI navigation agents and by extension every kind of open-ended task automation agent. What we're doing can be abstracted out into any agent that is trying to automate a complex task in the real world.

We choose GUI navigation because it's a much harder problem to solve than video games.  The web is a vast, open system, with pitfalls, errors, problems, good design, bad design and more.  It's a world of sparse rewards, where you have to click a lot of things and toggle a lot of switches to get a meaningful reward. 

In other words, it's hard and what we've learned there can help anyone trying to build an agent in a more controlled environment.

While the big AI labs are promising that we'll have models that can reason easily about any task and do it with very little business integration work and training, they are dead wrong.  For the foreseeable future, integrating these AI systems into the real world is the hardest problem, not to mention making them reliable, tying them into legacy systems and more.

This is not an open question. 

Most people just think it's an open question. 

They believe the big labs when they say the models will just be so good that there's nothing left for us to do.  Those folks have a bridge to sell you and they sold it to a lot of people but it's simply not true.

OpenAI has a blueprint for its five steps to AGI:

1. Level 1 ("Conversational"), AI that can converse with people in a human like way

2. Level 2 (“Reasoners”), means systems can solve problems as well as a human with a  doctorate-level education. We've seen the first stages of this with o1.

3. Level 3 (“Agents”) are systems that can spend several days acting on a user’s behalf.

4. Level 4 ("Innovators") are AI system that can develop new innovations and novel ideas.

5. Level 5 (“Organizations”), means an AI systems that can do the work of an entire organization, like a small or medium business

Each one of these steps is a quagmire in and of itself.  Each one has a series of mini-steps, setbacks and problems that are incredibly hard to overcome, not to mention painful integration points to make it all work in any single business or to accomplish a single task.

In other words, there is friction at every one of these steps. A lot of it.

At Kentauros we see an eightfold path to better agents:

1. Tool interface models

2. Parallel processing

3. Memory

4. Shared memory

5. Task specific reasoning

6. How swappable skills

7. Generalized task reasoning

8. Generalized reasoning

We don't see these as steps.

We see this as a major road with side roads, all of those roads branching and sometimes looping back to the other roads.

(Image Midjourney 6.1)

Of course, any project will need to have an order to tackle these in, because nobody can do these all at once. 

The good news is, once you go down one of these roads, it lets you start tackling other pieces of the stack.  It helps you understand the task you're facing, make mistakes, adjust and then move forward with what you learned.

Getting better at any one of these keys creates a positive feedback loop.

With better memory you have better workflow/tactical data which you can now loop back into your reasoning engine.