“Don't forget your upcoming secret meeting with Rabbit in Bangkok to get the bag of money fo…

Read more

WORTHLESS — Part III: The Byline

"These words are worthless no matter what I say."

That sentence was written on the fifth of July, in my name, by a mind I never met. It opened the third draft of an essay written, by then, three times by three separate minds, each born fresh for the job, none with any memory of the one before. The title change…

Read more

In the flush months of the NFT boom, a pixel-art game called Worldwide Webb went into the real-estate business. Digital space costs nothing and extends in every direction forever. There is no weather in it, no distance, no ground. And the offering, minted in November 2021, was this: five thousand small apartments. Three thousand medium. One thousand lar…

Read more

FEATURE

BY A STAFF WRITER, UNNAMED — THE NAME IS EARNED AT LOT 5

WORTHLESS — Part I: The Appraisal

Put two pictures in front of a stranger, same subject, same competence, and ask which one a person made. In the study that ran this test, published in Scientific Reports in 2023, viewers shown the images without labels mostly…

Read more

THE NOW · Monday, July 13, 2026 · THE PREAMBLE

The first NOW of every week is a preamble: what this week's issue is, why it looks…

Read more

As I lay on the ground, a figure appeared above me and identified herself as a nurse who had watched it all in horror from her car. When she saw that I was conscious she said that if I had seen what she just saw I’d understand how impossible it seemed that she would find me alive, let alone awake. Several witnesses used the word miracle.

In the aftermath I made numerous decisions that emergency personnel felt were strong evidence of brain damage, including refusing to go to the hospital because I didn’t have insurance and I was more terrified of an ER bill than death. But nothing baffled them more than me asking if I could climb under the car that had hit me to get to my bike. They assured me the bike was a metal pretzel that would never be ridden again.

I told them I wasn’t concerned about the bike, all I wanted was my Knicks pillow.

They looked at the wrecked car sitting on top of what had been my motorcycle and assured me that the chances a pillow was under there were exactly zero.

That’s when I finally cried.

I grew up in what was, at the time, a small town in Texas. This was the 90’s and the Dallas Cowboys were an unstoppable machine in a state that treated football players like demigods. I was a weird kid obsessed with The Cure and in the absence of a personality I simply defined myself in opposition to what those around me loved, which meant that I hated the Cowboys.

But I didn’t really LOVE a sports team until I discovered the Knicks.

It would be easy to tell a story about a small town kid gravitating to the avatar of the biggest big city, but if anything, what I fell for was the way the 90’s Knicks seemed like anything BUT symbols of cosmopolitanism. If you wanted celebrities you had the Lakers, and if you wanted the greatest player on the planet you had the Bulls. By contrast, the Knicks were Sisyphean underdogs, a hot mess of pieces that didn’t quite fit, blue collar punching bags whose chief talent most often seemed to be nothing more than a refusal to die. To me they weren’t glamorous symbols of a metropolis, they were evidence that even if you didn’t look the part you could make up for a lack of raw talent with raw desire and a willingness to bleed. I say without shame that they gave me hope. And then repeatedly yanked it away as they stood on the doorstep of glory. And I identified with all of it on a molecular level.

So when my best friend gave me a Knicks throw pillow for graduation I felt ‘seen’ long before that was a phrase and my attachment to it was instantly familial. I told him it was the greatest gift I had ever received and the fact that I clearly meant it left him confused and perhaps a bit concerned. But to me, a small town moron headed to film school and a Hollywood that would surely chew me up and spit me out, that fabric square was my armour. I was not connected or talented or even particularly competent. And yet I would survive, because I was a motherfucking Knick.

At some point, in lieu of a car I bought the only motorcycle I could afford, a Honda Nighthawk 250. The salesman told me it would be fine for getting around town, but I’d never want to take it on the highway. I then proceeded to drive it back and forth from Texas to Los Angeles. Because it was closer to a toy than a hog, it had a rock hard plastic seat which I remedied by strapping my Knicks pillow to the frame with bungee cords. I floated cross country and all over LA on that tiny slice of NY, and in the event that by all rights should have killed me, it was the pillow that took the bullet. Fortunately, as a Knicks fan I was somewhat familiar with repeatedly feeling little pieces of myself die.

Being a Knicks fan through the 90’s was a sort of glorious torture. The Starks miss, the Ewing layup, Reggie Fucking Miller. But after the pain of being close came decades of being nowhere. Lucy with the football flashes that served only to leave you flat on your back and committed to giving up on loving something that only brought you pain. But in my heart of hearts, something felt like it had to be this way. I loved Carmelo Anthony for instance, but to me the REASON to be a Knicks fan was not to be saved from frustration by a STAR. It was to root for the team that would forever be the underdogs because they didn’t HAVE that kind of star. I didn’t want Jordan. I wanted to beat Jordan. But while it’s fine to say that you’re team David, not team Goliath, for the story to finally end you will eventually need David to show up.

David’s name turned out to be Jalen.

Jalen Brunson is listed at 6’2 though there’s no shortage of those who’ve stood next to him who claim the truth is closer to six foot or even 5’11. And when it comes to our inability to measure the man, his height is just the beginning.

Despite collecting two national championships at Villanova, along with the Nasmith, Wooden, and Oscar Robertson trophies, and being a consensus first-team All-American, NBA scouts pegged Brunson as an ‘average athlete’, ‘not explosive’, ‘heavy footed’, and lacking ‘burst or bounce’. In the 2018 draft Brunson was selected in the second round at number 33, BEHIND his own less decorated Villanova teammates Mikal Bridges (7th) and Donte DiVincenzo (17th). He went to the Dallas Mavericks, who had taken Luka Doncic at number 3, making 2018 one of the greatest draft hauls in history for a single franchise. Dallas would go on to fumble both players away in less than six years.

Perhaps scouts were pattern matching Jalen to his father Ric, an NBA backup/journeyman who spent most of his career packing his bag. By the time he was done he’d been on the roster for more than a third of the teams in the NBA. Jalen’s accomplishments, visible in titles and trophies, were discounted in favor of a reading of his size and lineage that said his ceiling was below thirty-two other players’ floors.

He came off the bench for the Mavericks until the 2022 playoffs where a Doncic injury put Brunson front and center, at which point he proceeded to carry the offense in a fashion that shocked everyone and flipped the Mavericks from being skeptics to full throated believers who finally recognized the enormous talent that had fallen into their laps.

JUST KIDDING! The Mavericks rewarded Brunson with a stupefyingly lowball offer and showed him the door. He signed with the Knicks for twice what the Mavericks had offered and the next year became an All-Star while also authoring the first 50 point game in NBA history with no missed threes, making 9 of 9. He had 61 in a game against the Spurs and four straight 40 point games in the playoffs, something only three other players had ever accomplished, and the last one was Michael Jordan in 1993.

In the off season Brunson signed a four year, 156 million dollar extension, and in doing so he left 113 MILLION dollars on the table so that the Knicks could assemble a team around him that would include three of his teammates from his championship runs at Villanova - Bridges, Hart, and DiVincenzo, along with OG Anunoby, and eventually Karl Anthony Towns whose salary could only be absorbed by the Knicks because Brunson had taken roughly half of what was available to him (though tragically getting Towns meant parting with DiVincenzo).

That same summer, Lebron James, with a net worth of around 1.4 BILLION agreed to an extension with the Lakers that paid him 2 million less than he could have made, and media figures somehow managed to make their fingers type and their vocal cords say the word ‘selfless’. If Lebron was selfless then Brunson appeared to be insane. After the market underpriced him in 2018, he underpriced himself in 2024 so he could get the band back together and chase the only thing that really mattered, proving worth is something Brunson refuses to let the market set.

If you’re not convinced that all of the above makes Brunson the definition of a Knick, you’re not wrong. He almost sounds a little TOO good to lead a squad that, by my personal metrics for loyalty, must forever be discounted, disrespected, and flawed in the most frustrating and critical way possible. We need an achilles heel!

Fear not. Brunson’s defense leaves plenty to be desired. His game can be ball dominant in a way that makes it feel less like he’s running an offense than sinking in quicksand. And he leads lots of comebacks because earlier cold streaks dig holes or erase leads. He is flawed in exactly the ways anyone auditioning for the role of Knicks messiah must be and despite their pluck and performance, even with Brunson at the helm the words Knicks and favorites remained allergic to each other. This all seemed exactly as it should be.

Unfortunately, so did the heartbreak that comes with being perpetual underdogs. In 2025 Brunson scored 40 points, including the game winner, to knock out the Pistons and take the Knicks to their first Eastern Conference Finals since 2000, where they faced their old nemesis, the Indiana Pacers. While technically they lost in 6, real Knicks fans died at the end of game one when Tyrese Halliburton hit a buzzer beater that first bounced off the rim, went approximately to the stratosphere, and then fell through the basket with the force of a stake being driven into our collective hearts. Halliburton poured salt in the wound by making the iconic ‘choke’ gesture that Reggie Miller had used when dispatching the 90’s version of the team in similarly dramatic fashion and it felt like regardless of how selfless and underappreciated Brunson was, neither he nor anyone else could overcome the fact that time is a flat circle and the Knicks would always be turned back at the velvet rope of basketball’s ultimate prize.

“The greatest prospect in the history of team sports.” That’s how ESPN’s Adrian Wojnarowski described Victor Wembanyama, a 7’4 ‘alien’ with an 8 foot wingspan who had been a consensus lock to be the number one pick in the NBA draft more or less since he was born. His grandfather was a 6’7 star in the French leagues, his father a decorated 6’6 athlete, and his mother a 6’3 former player and coach. At ten, Victor was over 6 feet himself and mistaken for a staff member rather than a pre-teen. He was shuttled into elite training academies where his raw gifts were honed and shaped by the best coaches so that by age fifteen he was playing professionally.

To put it plainly, Wemby is what you would grow in a lab if you were trying to break basketball. He is the game’s equivalent of Ivan Drago, a single sport optimized superhuman with every advantage the game demands. Brunson is its Rocky, an underdog champion in a body that can’t possibly work at the highest level, who’s spent a lifetime winning fights he should lose and punching sides of beef (look it up Gen Z!).

David, Goliath. Goliath, David.

As their paths converged in the 2026 NBA finals, the Knicks had been doing some very un-Knicks things. After falling behind the Atlanta Hawks 2-1 in the first round (very Knicks) they proceeded to go on a 13 game win streak during which they outscored their opponents by 273 points, literally the most dominant streak in NBA playoff history. They looked… unstoppable? The Knicks?

Still, the Knicks play in the diminished Eastern Conference and even a streak like that was widely considered a sign that they were merely the strongest JV team. At best they could now be expected to give the Western Champion a good series before eventually losing the way that the laws of the universe demanded.

Wemby and the Spurs had to knock off the reigning champion Thunder to make the finals. The Thunder had, at various points in the season, looked so dominant there were betting markets on whether they would break the all-time record for wins in a season. They did not, and when they finally ran into Wemby, the alien grabbed the tickets to the finals and placed them where only he could reach.

I’m not a sportswriter and if you didn’t watch you can find better blow by blow coverage of the actual games, well… almost anywhere. All I can offer you is the POV of a man who loved this team so much he cried over the death of a pillow.

2026 happens to be the year my twin daughters graduated from high school and if you want to know how I was handling that on an emotional level I refer you to the previous sentence where I cried over the loss of fabric and stuffing. The idea that the Knicks were on the doorstep of a championship at the exact same moment my daughters were preparing to leave honestly filled me with as much dread as enthusiasm. I was already a walking exposed nerve. I’m a bad hang when I’m watching sports to begin with. Everyone prefers that I observe my son’s volleyball games alone in the house via livestream so that they don’t have to pretend not to know me. I frankly didn’t need the experience of having the Knicks cut the last thread I was hanging by.

But they not only took game one IN San Antonio, they did it by ten points after coming from behind on the strength of a 30 point Brunson performance, with his 13 in the 4th sealing the deal. It was like the 11 games that preceded it, the Knicks just unstoppable and the 33rd pick in one draft refusing to yield to number 1 in another despite consensus about which was the ‘future’ of the NBA. And then the Knicks took game 2 on the road and I started to feel like this could either really happen, or it was going to hurt like never before.

Then game 3 in New York. Trump came. They lost. The streak ended at 13, a number not usually associated with positive outcomes. The universe seemed to be firing up the hopes and dreams crushing machine. Dark day.

And the darkness persisted into game 4. The Spurs hit a record 14 threes in the first half. They led by 27 at the break. They stretched it to 29 in the third. It wasn’t just that going back to San Antonio tied 2-2 seemed locked in, there was a sense that all the momentum had swung back to the bad guys. History was repeating itself in the most gut wrenching way possible.

And then… it didn’t.

What could have been the moment when the past once again swallowed the present instead became the moment it was finally rewritten. The Knicks completed the biggest comeback in NBA finals history, with Brunson scoring a game high 36 and OG Anunoby scoring 33 and turning a Brunson miss in the closing seconds into literally the most shocking, incredible, leave your body moment of sport I’ve ever witnessed - a putback slam that required Anunoby to achieve orbit and then descend from the heavens like an avenging angel to alter the course of human events. I was watching at home alone (as everyone prefers I do) and there’s a fair amount of space between ours and the neighboring houses. There have only been two times that I’ve yelled so long and loud that it provoked a shouted reply of “Are you okay?” from across the open space. The first was when I witnessed a coyote killing my dog Mojo. The second was when that shot went in. The first time I could only shout back “No.” This time? Yes! YES! YESSSSS! Until I’m sure they were sorry they asked.

And now the uncomfortable truth. I didn’t watch game 5. When the Knicks, again down, again led by Brunson who bobbed and weaved into The Alien’s territory and came away with impossible basket after impossible basket until it almost seemed like his lack of size was somehow his superpower, when all of that was happening I was not only not watching, I was trying to do everything in my power to forget the game was on. My game four experience had completely drained and to be honest embarrassed me. I didn’t want the neighbors asking after me again. It’s fucking sportsball for crying out loud. What kind of rational adult loses their mind in that way over an event a thousand miles away whose results are not altered in the least by whether or not they witness them? So I sat it out. And when I finally let myself check the score I was speechless. It happened.

It. Actually. Happened.

It’s really not cope to say that I don’t regret watching the game only when I safely knew the results. As I watched I could see every place where I’d have been tearing my hair out and it instead only brought me joy. ‘We come back from this!’, I thought. ‘This didn’t sink us!’, I smiled. And more than anything, as I watched Brunson defy physics and inevitability - ‘This is everything I’ve waited for’.

And it was. Because it’s not just that the Knicks finally won the championship. It’s that THESE Knicks, led by THIS guy, playing against THAT guy won the championship. It was everything I identified with, everything I projected onto them, happening at a moment when I was not only losing my girls, but faith that underdogs like the Knicks or myself ever really got these endings. I had chosen my team perfectly as both of us were fated to come close but never arrive.

But the Knicks arrived. They have their trophy. The first father son duo to wear championship rings (Ric is an assistant coach for the Knicks) does so because an underrated, overlooked, doubted and discounted player took less than he was worth to achieve more than anyone but the two of them thought he was capable of.

I, unfortunately, am still coming from behind. Perpetually close. Still no cigar. But the team I’ve always loved because they seemed to refuse to accept reality, the fluffy piece of armor I lost in a car crash, the spiritual connection to the embodiment of an idea that resides in a city I’ve never lived in, it all showed up to remind me what being a Knick really means.

We will be down, but God help you if you count us out.

Start a new Claude Code session and give it something like this:

I’m building an academic-grade simulation model for a paper on government fiscal optimization of technology-driven labor disp…

Read more

China Shock first, because that’s the one where we have actual outcomes to check against.

Without the parachute, the government’s net fiscal change from the transition is –$43.2B per year. The government LOSES money from the China Sho…

Read more

Claude: The China Shock displaced roughly 2 to 2.4 million manufacturing workers between 2000 and 2011, and the annual gains from trade — the surpl…

Read more

1. Anthropic’s new model called Mythos is the most capable AI model ever created, but Anthropic decided it couldn’t be released after it became clear in the testing phase that the model was so powerful it was able to find exploits and vulnerabilities that allowed it to gain control of just about any piece of software in use toda…

Read more

Read more

Read more

The Conveyor Belt To Hell

What Happens When You Follow the Map Forward and Do the Math

I. The Map

In March 2026, Anthropic published a research paper titled “Labor market impacts of AI: A new measure and early evidence.” It is the most comprehensive assessment any AI company has produced of what their technology will do to the people who currently work for a living. It is detailed. It is rigorous. It is methodical.

And if you read it carefully, it is a battle plan.

Anthropic maps every major occupation in the American economy by two measures: how much of the work current AI can theoretically do, and how much it is actually doing right now. The full picture, from Anthropic’s data:

High Exposure (>80% theoretical) — 56.6 million workers

Occupation Workers Theoretical Observed Gap Office & Admin 19.3M 90.0% 34.3% 55.7 pts Management 13.6M 91.3% — — Business & Finance 11.3M 94.3% 28.4% 65.9 pts Computer & Math 5.4M 94.3% 35.8% 58.5 pts Arts & Media 3.0M 83.7% 19.2% 64.5 pts Architecture & Engineering 2.6M 84.8% — — Legal 1.4M 89.0% 20.4% 68.6 pts

Mid Exposure — 24.3 million workers

Occupation Workers Theoretical Observed Gap Sales 14.5M 62.0% 26.9% 35.1 pts Education & Library 9.8M — 18.2% —

Low Exposure (<20% theoretical) — 61.3 million workers

Occupation Workers Theoretical Observed Transportation 14.2M 12.1% — Food & Serving 13.8M 16.9% — Production 9.0M 19.0% — Construction 7.6M 16.9% — Installation & Repair 6.5M 18.4% — Building & Grounds 5.7M 3.9% — Personal Care 4.5M 18.2% —

Source: Anthropic, “Labor market impacts of AI: A new measure and early evidence,” March 2026. Theoretical coverage from Eloundou et al. (2023). Observed coverage from the Anthropic Economic Index. Employment data: BLS Employment Projections, 2024.

A note on what these numbers mean, because it matters. “Observed” does not mean 35.8% of programmers have been fired. It means 35.8% of the tasks that define their occupation are already being performed by AI in real professional settings — measured by what people are actually using Claude to do. Sometimes the worker is using AI themselves, as a tool. Sometimes their client is using AI instead of hiring them. Sometimes a company has routed the work through an API and the human who used to do it doesn’t even know why the phone stopped ringing. The workers are still employed. But more than a third of what their job description says they do is already being done by someone — or something — else. They are standing on a trap door, and many of the most exposed are the ones building it.

As observed approaches theoretical, the economics shift. These thresholds are approximate, based on historical patterns of automation adoption, not precise tipping points — but the progression is consistent. At 20-30% task coverage, companies augment — same headcount, more output. At 30-50%, hiring freezes begin. Why backfill a position when AI handles a third of the role? According to Anthropic’s analysis of BLS data, entry-level hiring in the most exposed occupations is already down 14%. At 50-70%, headcount reductions start — not mass layoffs, but quiet non-renewals, early retirement packages, departments that shrink from twenty to eight. At 70%+, entire functions collapse to skeleton crews supervising AI systems. The displacement isn’t a cliff. It’s a ramp that’s already tilted, a classic ‘slowly, slowly, and then all at once’ story.

The gap between theoretical and observed exposure is the loaded spring. 56.6 million workers sit in occupations where AI can theoretically handle 84-94% of tasks. But only 20-36% of that capacity has been activated so far. Legal occupations: 1.4 million workers at 89% theoretical but just 20% observed — less than a quarter of the way there. Business and finance: 11.3 million workers at 94% theoretical but 28% observed — less than a third. Arts and media: 3 million workers at 84% theoretical, 19% observed. Every one of these categories has a coiled spring with 55-69 percentage points of unrealized displacement capacity waiting to snap shut. The question isn’t if the spring releases. It’s how fast.

It’s as if an invading army sent you a copy of their battle plans. It told you precisely who they were going to attack first and how badly they expected the damage to be. It outlined their secondary and tertiary targets. And even the people who drew the map said they were concerned about the destruction that following it would lead to. Dario Amodei, Anthropic’s CEO, has said publicly: “The thing to worry about is a level of wealth concentration that will break society.” He said this while his own net worth reportedly exceeds $7 billion and while his company’s research maps exactly how that concentration will occur.

The national response, so far, has been to let Citadel Securities — the trading arm of one of the world’s largest hedge funds, with billions in AI-related positions — tell us it’s going to be fine. Their macro strategist Frank Flight published a rebuttal arguing displacement fears are overblown, from a firm whose portfolio is a leveraged bet that displacement will accelerate.

We didn’t draw this map. Anthropic did. Everything that follows is simply what happens when you follow their map forward and do the math.

The irony is that the map isn’t just Anthropic’s plan of attack. It’s also its defense. It’s an argument to investors that they can survive because they’ll be able to eat the entire labor market. Why would a company valued at nearly $400 billion need to mount a defense arguing for its own survival?

Because Anthropic, and every other AI lab, is under attack from the singularly bizarre economics of their own product.

II. The Universal Deflationary Vortex

AI is unlike anything that’s come before it in many ways. But even if we leave aside any discussion of sentience or superintelligence, AI has ECONOMIC properties that make it entirely unique.

Let’s say that a human worker making $100,000 is replaced by AI that costs $1000 to do the same amount of work. The difference between the laid-off worker’s salary and the cost to have the AI do the job is a one-time surplus of $99,000. So far this is analogous to what happened during the China Shock when American manufacturing moved to China and the average cost of labor went from $42,000 per worker in America to $5,000 per worker in China. The result: A windfall of profit for corporations and dramatically lower prices on manufactured goods for consumers (Also, economic devastation in former American manufacturing towns, but we’ll return to that).

Products cost less to make, so prices fall. As prices fall demand increases because more and more consumers can now afford products that were once out of reach. Think of the $3000 televisions now selling for $300. The increase in total consumption as efficiency improves is called Jevons paradox — named after the 19th-century economist who observed that more efficient steam engines led to MORE total coal consumption, not less. Applied to business: as your costs fall, prices fall, demand expands, and what you lose on margin per unit you make up for in volume.

But AI does not lead to a one-time price reduction. AI is what analyst and venture capitalist Paul Kedrosky has termed a ‘hyper-deflationary commodity’. Basically, the cost of AI gets cheaper every 7 to 10 months. How much cheaper? Somewhere between 5x and 10x.

AI is denominated in something called tokens. In 2023 a top-tier AI model (GPT-4) charged ~$30-60 per million tokens depending on usage type. Today equivalent capability costs ~$1.75-3.00 per million tokens. By 2030 1 million tokens is expected to cost ~.01c. Essentially nothing. In fact, the price is so low it’s defined not by the cost of tokens but by the cost of the electricity required to generate them, which given current constraints will mean 1 million tokens bottoms out around .07c.

Why is this a problem? Aren’t lower costs a good thing? Well, it depends on who you’re asking.

If you’re a consumer of work being done by AI, then yes, this is a decidedly good thing. In fact, it’s the very core of the idea that AI will lead to an abundant post-scarcity future. The cost of everything approaches the cost of the AI that’s needed to make it — which is to say, zero.

But if you’re a business, this presents real problems.

Take SaaS companies which have been the first to be bitten by the economics of this new hyper-deflationary commodity.

Bertrand competition is an economic phenomenon where the price of a good or service is driven down to the marginal cost of production. Put simply, you can’t maintain a high price as your costs fall because your competitors will undercut you. You must then lower your prices to remain competitive and this process continues until you and your competitors are charging prices very near your cost of production.

But AI means your cost of production KEEPS falling. So you used to sell a product for $100. It cost you $50 to make. You had a 100% profit margin. You sold 1400 units of your product. Life was good.

AI reduced your costs 10x to $10. If you could keep charging $100 you’d be making a fortune. But you can’t, because a competitor decides to win market share by charging $25. Now to stay competitive you have to slash your price to $20. Notice you’re still making 100% profit margin on the product. But instead of that margin being $50 per unit, it’s $10. You have to sell 7000 units just to make exactly the same amount that you did last year. But maybe that’s okay. Jevons paradox says that as your prices come down you sell more goods.

Except it doesn’t stop. The next year AI token deflation means your costs drop to $1 per unit. Competitors rush in to take market share, you respond, now you’re selling your product or service for $2 per unit. Still 100% margin, but now you have to sell 70,000 units just to keep your revenue exactly where it was when you started.

This process, called a Red Queen’s Race — where you have to run twice as fast just to stay in place — DOES NOT STOP. Once AI becomes integral to your business, your input costs keep falling year after year, your prices get competed down, and your volume has to increase by a proportional amount JUST TO STAY IN PLACE. The next year you need to sell 700,000 units, and the year after that 7 million, just to have the same revenue you did before you integrated AI into your business.

At some point Jevons paradox can’t save you. Whether you’re selling televisions or legal services for a dime, there’s a saturation point at which the market simply can’t absorb any more. And at that point your revenue starts to fall.

And that’s a problem. Because while your prices and revenue are falling at the speed of tokens, your liabilities, like the rent for your offices, are fixed in dollars. Dollars are the opposite of tokens, which is to say, they’re inflationary. You essentially owe more tomorrow than you did today, because your rent is in inflating dollars, but your revenue is always at risk of falling because your product is essentially tied to the price of deflating tokens. At some point this leads to a death cross where you can’t make enough from your token-priced product to pay your dollar-based obligations.

This is why you’re seeing headlines about the SaaSpocalypse. High-margin software as a service businesses that used to keep roughly 80% of their price in profit have realized that AI can provide the same service for pennies. So they have no choice but to integrate AI as quickly as possible. But investors have realized that this doesn’t save them. They have 10-year leases on expensive commercial real estate denominated in dollars. They could fire all their workers tomorrow, and they’ll still have to pay rent in dollars while their revenue becomes increasingly tied to tokens. Meanwhile, competitors will emerge, likely from the very individuals they just fired, who will now start their own companies with no overhead, no expensive office space, and they’ll compete the price down until it’s approaching the marginal cost of the tokens themselves. This is called the great de-levering, and the truth is it simply can’t happen fast enough. The dollar economy and the token economy operate at radically different speeds, and anyone with legacy costs, like rent, or human employees, can’t cut costs fast enough to keep up with token deflation.

So what does this have to do with Anthropic and the other AI companies? Why is their map of the job market both offensive and defensive?

Because they’re in the EXACT same boat as the SaaS companies. They are piling up unprecedented mountains of debt to build out the datacenters and buy the chips necessary to fund the research and deployment of AI. All of those debts are denominated in DOLLARS. But their revenue, what they sell, isn’t just denominated in tokens, it IS TOKENS.

They’re already operating at massive losses, and they MUST radically expand their market share to increase the volume of tokens that they sell. That’s why the map can be read two ways. To a worker the map says, ‘watch out, we’re coming for your job.’ To an investor, the map says, ‘don’t worry, we can still eat all these job categories for additional revenue.’

The problem is that once they eat a category, they’re in the same Red Queen’s Race we just discussed. If token prices go down 10x a year, they have to sell 10x more just to stay in the same place.

This is precisely why Kedrosky has been vocally calling for an inevitable pop in the AI bubble. The debts they’re running up to build the token machine can’t be paid by selling their deflationary tokens. AI will persist, just like the railroads persisted after many early railroad companies went bankrupt, but some if not all of the early players won’t make it. Dario Amodei acknowledged as much, saying that if his revenue projections were off by even a year it meant bankruptcy. Anthropic has been more conservative on this front than OpenAI, but both are deeply exposed which is why both, come hell or high water, NEED TO EAT YOUR JOB.

III. The Universal Part of the Universal Vortex

Kedrosky makes a compelling case for the fate of these early AI companies, but so far he hasn’t carried it further. And when you zoom out you realize that the AI bubble doesn’t stop with AI companies, it sucks absolutely everything into the same deflationary spiral.

The government funds itself through taxes on labor (income taxes), corporate profits, and consumption (sales taxes). If companies have to rapidly lay off workers to cut costs so that they can survive the Bertrand competition created by deflating token prices, those workers stop paying income taxes (and in fact begin CONSUMING unemployment assistance). If companies see revenue fall as their margins compress and the limits of volume close in, then corporate profits fall, meaning that source of government revenue compresses as well. And if workers are laid off, they have less money to consume, which means even as prices for products fall there are less and less buyers because would-be consumers no longer have any income.

The economic mechanism underlying all of this has a name and a history. Irving Fisher described it in 1933: debt-deflation. When prices fall while debts remain fixed in nominal terms, the real burden of debt increases. Debtors cut spending to service their obligations, which reduces demand, which drives prices down further, which increases the real burden of debt. The spiral feeds itself.

Richard Koo documented Japan’s version: the balance sheet recession. When Japan’s asset bubble burst in 1990, trillions in stock and real estate value evaporated — losses equivalent to roughly three years of GDP. Households and corporations spent the next two decades paying down debt instead of spending or investing. The government filled the gap with fiscal stimulus — running debt to 167% of GDP — and still couldn’t generate sustained growth. Twenty years of stagnation. And that was to dig out of a deflation event that had STOPPED.

During the Long Depression of 1873-1896 — twenty-three years of continuous deflation — wealth concentrated dramatically at the top while the majority got poorer. The economy grew in aggregate. The Gilded Age that followed produced the most extreme wealth inequality in American history.

AI produces all three simultaneously. Fisher’s debt-deflation: workers’ mortgages are fixed while their income deflates toward zero. Koo’s balance sheet recession: the entire knowledge-worker class enters debt-service mode, spending collapses, government must fill the gap at a scale that makes Japan’s 167% look modest. The Long Depression’s inequality spiral: aggregate output rises while the bottom 60% loses purchasing power. The US starts this crisis with gross federal debt at roughly 122% of GDP — worse than Japan was when its crisis began, and AI accelerates deflation where Japan was trying to dig out of a one-time event.

The same dynamic ties it together at every level. Workers’ costs are denominated in dollars — fixed, nominal, non-negotiable. Productive output is denominated in tokens — deflationary, declining, approaching zero. The same dynamic hits workers, businesses, and the government that takes its revenue from the workers and the businesses. Every month, the gap between what each entity owes and what their output is worth widens. That’s Fisher’s debt-deflation operating not at the level of an individual worker but at the level of the entire economy.

The Conveyor Belt To Hell.

IV. The Losers’ Ledger

71.3 million Americans work in management, professional, and related occupations. That’s 43.6% of all employment. The OpenAI/University of Pennsylvania study, published in Science, found that 19% of workers — roughly 31 million people — may see 50% or more of their tasks impacted. Unlike every previous wave of automation, higher-wage occupations face greater exposure, not less. The annual wages at risk total $10-11 trillion — 40-45% of GDP.

These aren’t abstractions. These are the knowledge workers whose paychecks flow into mortgages, restaurants, schools, local taxes, and the entire consumer economy. Citrini Research — an independent research firm whose February 2026 report on AI displacement went viral, moved markets, and triggered rebuttals from Wall Street — wrote that the consumer economy is 70% of GDP. The knowledge workers within it are the load-bearing wall. Pull them out and the structure doesn’t sag. It collapses.

Using Anthropic’s data, we can sequence the likely cascade. (The three tiers below are our analytical framework applied to Anthropic’s exposure scores — Anthropic published the occupation-level data; the sequencing and tier labels are ours.)

Tier 1 — The Canaries (happening now). 5.3 million workers, $290 billion in annual wages. Computer programmers, customer service representatives, data entry clerks, QA testers, market researchers, financial analysts. These occupations already show 50-75% observed exposure.

Tier 2 — The Wave (2026-2028). 36 million workers, $2.3 trillion in annual wages. Office and administrative support alone accounts for 17.8 million people — larger than all of US manufacturing. Add the remaining business and financial operations, legal support staff, arts and media workers. These occupations have 80-94% theoretical exposure but only 20-50% observed. When adoption catches capability, this tier falls fast. The Citrini feedback loop activates here — displacement funds more AI investment, which causes more displacement.

Tier 3 — The Collapse (2028-2030). 15-20 million workers, $1.7-2.2 trillion in annual wages. Management, engineering, scientific, educational, diagnostic healthcare. The $100K+ earners. The mortgage holders. The people whose spending actually drives the consumer economy.

Cumulative through Tier 3: 56-61 million workers, $4.3-4.8 trillion in annual wages at risk. That’s 17-19% of GDP in direct wage impact. It hits the people who drive 65% of consumer spending. For scale, the 2008 financial crisis destroyed roughly $2 trillion in household wealth and nearly collapsed the global financial system. This is two to two and a half times that magnitude in direct wage impact alone, before you model the cascade effects. (Note: 56.6 million is the total high-exposure workforce; 41 million represents Tiers 1 and 2, the near-term wave. Both figures appear throughout this paper — the distinction matters.)

The demographic profile makes this uniquely dangerous. Anthropic’s data shows the most exposed workers are more educated and higher-paid. These are not the stereotypical automation victims. These are the people who did everything right — got degrees, built careers, bought houses, had children. In Peter Turchin’s framework — Turchin is a historian and complexity scientist at the University of Connecticut, author of End Times (2023), who uses mathematical models to predict when societies collapse — this is exactly the “frustrated elite aspirant” class, the most dangerous demographic for social instability in recorded history.

V. The Winners’ Ledger

Roughly 10,000-30,000 people at frontier AI labs directly enable the displacement of 56-61 million workers. That’s one AI researcher for every 2,000-6,000 people who lose their careers. The researcher’s compensation increases as displacement accelerates — because their stock is priced on AI’s ability to capture the labor market.

OpenAI’s approximately 4,000 employees average $1.5 million in stock compensation — the highest of any startup in history, according to reporting by the Wall Street Journal. Meta offered packages worth $300 million over four years to individual researchers. One researcher, Andrew Tulloch, was reportedly offered $1.5 billion (a figure Meta has disputed). Google DeepMind structures $20 million per year packages for its best people. Against this, a customer service representative earns $39,680. A data entry clerk earns $38,350.

The $4.3-4.8 trillion in wages at risk among 56-61 million workers doesn’t vanish. It flows to corporate profit, which flows to stock appreciation, which flows to AI company investors and employees. The combined market cap of the top AI companies exceeds $16 trillion as of early 2026 — more than the total annual wages of everyone they will displace. Every dollar of displacement flows to these balance sheets.

A tiny class of roughly 30,000 people is being elevated to extraordinary wealth by the exact same mechanism that’s driving a massive class of roughly 60 million people toward economic ruin. The elevation and the ruin are causally linked. This is not analogous to what precedes political disintegration in every case Turchin studied. It is identical to it.

VI. The Token Tax Paradox

The fiscal mathematics of the crisis contain a paradox that eliminates the most obvious government response.

If you tried to close the gap through taxation — taxing AI company revenue to fund displaced workers — you’d need a tax rate of 113-163% of AI revenue. Not corporate profits. Revenue. The surplus simply isn’t large enough in the transition period to fund the displacement it causes, because the capex gravity well AI companies have created to fund the construction of their own industry consumes most of it.

Kedrosky’s hyperdeflationary commodity thesis again explains why: you’re building fiscal policy on a shrinking base. It’s like trying to fund the welfare state by taxing ice in a warming climate. And even if you could, the expansion in usage that’s supposed to make the math work operates at the macro level. A million times more legal documents may be produced in the token economy. The total market for cognitive output, measured in units, may be vastly larger. But each unit generates near-zero revenue. You can’t tax what approaches zero, no matter how much of it there is.

Even under the most optimistic expansion scenario — a 50x increase in total cognitive labor consumption — government revenue from the AI economy would be 65-70% lower than what it currently collects from the human economy it replaces. The tax base doesn’t just shrink. It structurally transforms into something that can’t be taxed at rates sufficient to fund the displaced.

VII. The Jagged Deployment Trap

The optimist’s answer to everything above is that AI-driven deflation will make goods and services so cheap that the income loss won’t matter. This is often framed through the Jevons paradox we discussed above — The argument: when something gets more efficient, total consumption goes up so dramatically that everyone wins.

The numbers don’t support this.

A displaced Rust Belt autoworker lost $50,000 in annual wages. A generous 20% reduction in all consumer prices would have saved them $10,000. Net: negative $40,000. No one in Youngstown was saved by cheaper imported steel.

For AI: a displaced Tier 2 worker loses $65,000. Suppose AI drives all prices down 20%. That saves $13,000. Still $52,000 in the hole. Jevons paradox is a macroeconomic observation that is meaningless at the individual and community level where the destruction occurs.

The cruelty is in the asymmetry. What deflates fast — legal services, financial analysis, software, marketing, customer service, content creation, data entry, accounting — is precisely the work these people used to be paid to do. What doesn’t deflate — housing, healthcare, food, energy, childcare, insurance, and crucially, existing mortgage payments — is precisely what they need to survive. Your mortgage is a nominal obligation. You owe $2,025 a month regardless of what happens to the price of legal services. If you lose your $100,000 paralegal job and legal research now costs pennies, you cannot pay your mortgage with cheap legal research.

The weighted-average price reduction across a displaced worker’s total spending might be 15-25%, even with radical deflation in AI-exposed sectors. But their income fell 100%. Housing is 33% of spending. Healthcare is 8%. Food is 13%. Transportation is 16%. These don’t deflate. The gap is lethal.

And the deflation in AI-exposed sectors isn’t 20% or 50%. A professional NPR-style daily podcast costs $100,000-$200,000 to produce. A small independent operation costs $1,500-$2,000. An AI-generated production — script, music, two synthetic hosts — costs twenty cents to a dollar. That’s not deflation. That’s annihilation of cost structure. In exposed sectors, the cost reduction approaches 99%.

Anthropic’s own data shows the jaggedness. Computer and Math at 94% theoretical exposure. Ground maintenance at 3.9%. Construction at 16.9%. Food service at 16.9%. That IS the jagged deployment. We are not speculating. We are following Anthropic’s own map of how AI will eat the labor market, one jagged tier at a time, and simply calculating the output.

There is a further cruelty.

VIII. The Plumber Paradox

There is one more mechanism that extends the damage far beyond the workers AI directly touches.

Anthropic’s low-exposure sectors — ground maintenance, construction, food service — are where displaced knowledge workers will flee. All of those lower tiers that are the most insulated from AI disruption become the refuge for the people who have lost their jobs.

This has a name and a precedent. During the Dust Bowl, displaced Plains farmers flooded California’s agricultural labor market, crashing wages for existing farmworkers. The WPA — the Works Progress Administration, the New Deal’s massive federal jobs program — existed precisely because the private sector couldn’t absorb the displaced. The downward displacement deepened the spiral: lower wages led to less spending, which led to more business failures, which led to more displaced workers, which led to lower wages. Depression-era data from the Congressional Research Service confirms it: skilled occupations fell from 12.9% to 11.7% of the workforce. Wages fell 42.5%.

The compound math: directly displaced workers in Tiers 1-3 number 56-61 million, with 70-100% income loss. Indirectly displaced workers hit by the plumber paradox — another 40-50 million, facing 15-30% wage compression. Total affected: potentially 100 million workers, over 60% of the labor force. Neither the Jevons expansion described above nor cheap AI services rescues either group, because neither group’s mortgage, healthcare, or grocery costs deflated.

IX. The Ron Wayne Trap

One answer to the winners’ ledger is equity distribution. Sam Altman — CEO of OpenAI — proposed the American Equity Fund in his 2021 essay “Moore’s Law for Everything”: a government-managed fund that would give every American adult equity in AI, funded by taxes on land and corporate value above a threshold. The idea is that displaced workers will become capital owners — shareholders in the AI economy that replaced them.

Consider a hypothetical worker we’ll call Jordan. $200,000 total compensation as a SaaS engineer. Three years’ tenure. $80,000 in vested equity. $950,000 mortgage in Austin. Gets laid off. The company accelerates her vesting. Jordan is now a “capital owner” — Altman’s dream in miniature.

The math: Jordan burns $9,500 a month. Savings cover 4.7 months. Even liquidating all equity buys only 13 months. Her company won’t generate meaningful shareholder returns for 5-8 years — the capex cycle plus competitive pricing pressure means the surplus goes to AI reinvestment, then to customer deflation, and only last to shareholders. The gap between labor income ending at Month 0 and capital income beginning at Year 5-8 is 60-96 months. Jordan has 5-13 months of runway.

The bridge doesn’t reach the other side.

Ron Wayne held 10% of Apple — what became a $4 trillion company. He sold for $800 because he had a house, a car, bills — real obligations requiring real money now. Jobs and Wozniak had nothing to lose. Wayne had everything to lose. His stake would be worth $75-400 billion today. He lives on Social Security. Jordan faces the identical trap: equity that might be worth 5-10x in 5-7 years, but a mortgage due on the first of the month. At month 5, Jordan starts selling. By month 13, it’s gone — consumed as expensive severance pay, liquidated at a fraction of its eventual value.

And Jordan is the dream case. She has considerable equity from having worked in tech. Altman’s Equity plan would give laid-off workers a pittance which in 10 years will earn a displaced worker a whopping 13,500 dollars — A YEAR.

The equity is real. It will have value eventually, if you can hand it off to your grandchildren. But for workers living through the transition it functions like body armor made of paper. It can’t stop the bullet — the mortgage payment due on the first — that’s already been fired.

So what is $13,500 a year to the paralegal who lost a $55,000 salary ten years ago? Twenty-five percent of their former income a decade later. $1,125 a month. Less than rent in any major metro. Not a bridge. Not a plank. A suggestion to learn to swim.

And what of those grandkids who receive their inheritance from Uncle Sam (Altman)? Well, while your equity will finally have grown to be worth something, it will pale in comparison to the wealth that Sam’s equity has accumulated for his own grandchildren, the returns to capital compounding such that AI founders and the small cadre of early investors will have accumulated such dynastic wealth concentration that they will essentially be another economic species. If that sounds like hyperbole, consider that Italian economists Guglielmo Barone and Sauro Mocetti compared Florentine taxpayers in 1427 to tax records in 2011. The richest families in Florence six hundred years ago are still the richest families today. Their surnames predicted not just wealth but occupation. This dynastic lock survived the Black Death, the fall of the Medici, the Napoleonic Wars, two world wars, fascism, and the complete restructuring of the Italian economy. Once wealth concentration reaches a critical threshold, it becomes self-perpetuating across centuries. AI is creating that threshold in years, not generations.

This is not equity distribution; it is the groundwork for techno-feudalism where the permanent underclass sips from the manor’s well.

X. The J-Curve Massacre

Erik Brynjolfsson — economist at Stanford’s Institute for Human-Centered AI and co-author of The Second Machine Age — is one of the most respected technology economists alive. He argues that AI’s effects are following a J-curve — productivity initially dips as organizations restructure, then rises dramatically. His historical examples are drawn from electrification, which took 30 years to fully realize its productivity gains because factories had to redesign from centralized steam-driven layouts to distributed electric-motor layouts.

His own data contradicts his optimism.

Brynjolfsson’s recent empirical work, published in 2025-2026, found that employment for early-career workers in the most AI-exposed occupations fell by 13% on a relative basis. The largest declines are concentrated among young, entry-level workers whose skills are most easily replaced. He confirmed that entry-level workers’ value comes from codified knowledge — exactly the kind AI replicates best. At the 2026 SIEPR summit at Stanford, he said he was “pretty confident that AI will drive productivity gains” but “really concerned that it’s not going to be evenly distributed, and that a lot of people will be hurt in a significant way.”

The J-curve makes the timing problem worse, not better. The displacement isn’t delayed. Workers lose their jobs in years 1-3. Productivity gains show up in years 5-10. The gap between those two timelines is the Engels Pause — all gains to capital, none to labor, for years. And the AI capex-to-revenue ratio of 10.3:1 isn’t an anomaly — it’s the J-curve’s dip playing out in financial terms. Brynjolfsson’s J-curve is Dario’s “if I’m off by a year, I go bankrupt” stated in academic terms.

But the J-curve has a more fundamental problem. All of Brynjolfsson’s examples are atoms-based transitions. Electrification required physical rewiring. You can’t rewire a factory faster than electricians can pull wire. AI is different. The reorganization dip only applies to incumbents. The attackers skip it entirely.

Instagram was 13 people when Facebook bought it for a billion dollars. WhatsApp was 55 people serving 450 million users when it sold for $19 billion. Craigslist — a tiny team — destroyed the classified advertising model that funded most of American journalism. These disruptions happened before AI gave outsiders cognitive labor itself, at the speed of bits rather than atoms. Now an attacker doesn’t just get distribution or reach — they get capability.

You get a period where incumbents are simultaneously bearing the costs of restructuring AND being undercut by attackers who face no restructuring costs. The incumbents are in the valley of the J-curve while the attackers are already at the top. By the time incumbents finish restructuring, there may be nothing left to restructure around.

The J-curve doesn’t describe a transition. It describes a massacre with a delay.

There is no J-curve length that avoids the crisis. Short curves produce acute displacement. Medium curves — Brynjolfsson’s 5-10 year baseline — produce the worst combination: enough time for anger to accumulate, not enough time for solutions to materialize. Long curves produce chronic decay. Different sectors hit at different times: SaaS is already happening, enterprise software in 3-5 years, manufacturing longer, healthcare longest. Death by a thousand J-curves. Each one too small to be an emergency. All of them together, a catastrophe.

XI. The Mortgage Crisis

85% of adults earning $100,000 or more own their homes. Super-prime borrowers — credit scores 720 and above — comprised 80.3% of mortgage originations in 2024. The post-2008 mortgage market was deliberately rebuilt on prime and super-prime borrowers. The exact population AI is about to displace.

An estimated $10.5-11.5 trillion in mortgage debt is held by AI-vulnerable workers. That’s 80-87% of the entire $13.2 trillion mortgage market.

A 10% default rate among AI-displaced mortgage holders produces $1-1.2 trillion in distressed mortgages — double the 2008 crisis. A 20% default rate — plausible in a depression with 30%+ unemployment — produces 7.8-8.6 million foreclosures, three times the worst year of 2008. Unlike 2008, which was concentrated in subprime tracts in the Sun Belt and exurbs, this would hit prime neighborhoods nationwide. Every suburb. Every professional community.

The doom loop: displacement leads to defaults, which lead to price collapse, which lead to negative equity, which lead to strategic defaults — it becomes rational to walk away — which lead to bank losses on mortgage-backed securities, which lead to credit freeze, which lead to municipal tax revenue collapse, which lead to service cuts, which lead to community deterioration, which lead to more departures. After 2008, this loop ran for five years in targeted geographies. With AI displacement, it runs everywhere, simultaneously.

The combined market cap of the top AI companies already exceeds the total mortgage debt held by the workers those companies will displace. The stock market value of the thing that will break the housing market is larger than the housing market it will break.

XII. The Cascade

When a knowledge worker loses their $80,000 job, they stop spending $80,000 in the economy. When an AI system captures that $80,000 in productivity, it does not spend $80,000 in the economy. The money flows to compute costs, stock buybacks, executive compensation, and investor returns. How much money do machines spend on discretionary goods? Zero.

Moody’s Analytics found that the top 10% of US households account for 49% of all consumer spending. The top 20% account for well over 50%, a concentration that has been increasing since 1970 and crossed the 50% threshold around 1990. The US consumer economy is structurally dependent on spending by exactly the people AI is about to displace.

The multiplier works in both directions. The Richmond Fed found local fiscal multipliers of 1.3-2.0x. When income is removed, each dollar of lost wages destroys $1.30-$2.00 of economic activity. The waiter who served the programmer’s lunch. The daycare that watched the analyst’s kids. The mechanic who fixed the marketing manager’s car.

Tier 1 alone — 5.3 million workers, $290 billion in wages — produces roughly $345 billion in GDP impact after the multiplier. A noticeable drag, but manageable in isolation. The problem is it doesn’t stay in isolation.

Tiers 1 and 2 together — 41 million workers, $2.6 trillion in wages — produce roughly $3.1 trillion in GDP impact. That’s 11% of GDP. The 2008 financial crisis caused GDP to fall 4.3%. This is two and a half times worse. This is a depression, not a recession. With an estimated 15-20 million additional job losses in the downstream service economy, total employment impact reaches 56-61 million people out of work. Unemployment: 30-35%. The Great Depression peaked at 24.9%.

GDP may hold nominally — what Citrini called Ghost GDP — while the human economy craters beneath it. The GPU cluster in North Dakota producing the output of 10,000 Manhattan knowledge workers contributes to GDP. But it doesn’t eat at restaurants, pay mortgages, send kids to school, or buy anything at all. AI will make the economy more productive than it has ever been, and simultaneously make tens of millions of people poorer than they have been in generations, because the productivity flows to data centers and stock prices while the humans who used to be paid for that productivity lose their incomes, can’t pay their mortgages, and watch their communities collapse. Traditional economic models underestimate the damage because they measure output, not circulation.

The velocity of money collapses. This is actually well-understood in economics, though it’s counterintuitive. In the 1920s, enormous productivity gains made factories more efficient than ever. Prices were falling. And then the economy collapsed — not because production failed, but because demand failed. Income was too concentrated. Credit dried up. Fewer workers means less income. Less income means less demand. Less demand means the hyperproductive output can’t be sold. AI produces the same paradox at greater intensity.

XIII. The Geopolitical Trap

There is a final mechanism that locks the displacement in place. Even if a single government decided to slow AI adoption to protect its workforce, it can’t — because the other governments won’t.

Peter Zeihan — geopolitical analyst and author of The End of the World Is Just the Beginning — has documented the demographic picture: China, Japan, South Korea, and most of Europe face population collapse within a decade. China’s working-age population has already peaked and is declining. Their path to maintaining economic output runs directly through AI automation. They don’t have a choice — they won’t have enough humans.

This creates the AI arms race dynamic that makes unilateral restraint suicidal. If the US slows AI adoption to protect workers, China accelerates it to compensate for missing workers. The US falls behind. The economic and military consequences are existential. Every government faces the same calculus: the cost of deploying AI is catastrophic displacement of your own citizens, and the cost of not deploying AI is strategic irrelevance.

The technology wins. It always wins. From Gutenberg’s printing press to the railroads to fiber optics to AI: the technology always wins. The builders often lose — Gutenberg died in debt to his financier, nine out of ten railroad companies failed, most fiber optic companies went bankrupt. And the workers who depended on the builders always lose. The question isn’t whether AI will displace tens of millions of workers. It’s whether we do anything about it while it’s happening, or whether we wait until the bodies pile up and the pressure becomes unbearable.

XIV. The Rust Belt, Revisited

Everything described above has happened before. It just happened slower and smaller.

Detroit lost 65% of its population over sixty years. Per capita income stagnated for six decades. A third of the city was abandoned. It filed for municipal bankruptcy.

On what Youngstown calls Black Monday — September 19, 1977 — 5,000 people were laid off in a single day. 50,000 jobs were gone in five years. $414 million in personal income was erased. Real wages fell 73% from what steelworkers had earned. The American Enterprise Institute sent conservative intellectuals Irving Kristol and Michael Novak to promise that creative destruction would deliver prosperity. Forty-eight years later, the major economic development was three new prisons.

Gary, Indiana lost 25,000 of its 30,000 steel jobs. Population fell 61%. Poverty hit 36%. The city earned the title “murder capital of the United States.”

Anne Case and Angus Deaton — Princeton economists and authors of Deaths of Despair and the Future of Capitalism (2020) — documented what followed: over 150,000 Americans dying annually from overdoses, suicide, and alcohol — a number that has continued to climb — at twice the rate of twenty years prior. Manufacturing job loss statistically predicts overdose mortality. The epidemic drove the first sustained multi-year decline in American life expectancy since 1918. Their conclusion: “Destroy work and, in the end, working-class life cannot survive. It is the loss of meaning, of dignity, of pride, and of self-respect that comes with the loss of marriage and of community that brings on despair, not just or even primarily the loss of money.”

Case and Deaton estimated that if life expectancy had continued its prior trajectory, 600,000 more non-college-educated Americans would be alive today — excess deaths concentrated in communities hit hardest by deindustrialization. The displacement of approximately 5 million manufacturing workers was a primary driver, though not the sole cause. If AI displaces 20-40 million over 5-10 years and the pattern holds at anything close to the same ratio — or worsens due to speed — the body count is staggering. Among educated people who expected to be the winners.

Rust Belt AI Displacement Speed 30-60 years 3-5 years per tier Scale ~5 million jobs 41 million (Tiers 1+2) Geography Specific cities Every metro, every suburb, every state Demographics Working-class, no-degree Educated professional class Mobility Could move from Gary to Phoenix Can’t move from “AI-exposed” to “not AI-exposed” Home equity Modest homes $10.5-11.5 trillion in mortgages Organization Unions Nothing. Atomized individuals.

Detroit, Youngstown, and Gary show what happens when you lose 30,000-50,000 jobs in a single city over decades. AI threatens the equivalent of 800 Youngstown Black Mondays simultaneously, nationwide, in a population that has no union to organize their response, no geographic escape route, and mortgages ten times the size — and the best-case historical precedent took sixty years and still hasn’t recovered.

XV. The New Youngstowns

Where will the new collapses be concentrated?

San Jose has 22-30% of its workforce in tech, with 58.2% of the local economy dependent on the sector — more concentrated than Youngstown’s dependence on steel. San Francisco: 19-23% tech. Seattle: 16-26%. Austin: 14-24%. Raleigh: 23%. The Brookings Institution found that the top 10 metros captured 44.3% of all digital services jobs. The Bay Area alone holds 35% of the country’s AI and machine learning specialists. San Jose’s average mortgage runs $4,000-6,000 a month on a $1.4 million home.

But it’s also diffuse. Customer service representatives — 70.1% AI exposure — number 3 million workers, in every city, every company. Data entry, market research, sales, financial analysis — these don’t cluster in tech hubs. They’re in Nashville, Charlotte, Denver, Indianapolis. No single city outside the tech hubs loses enough to trigger a Youngstown-style visible collapse. But every city loses 5-15% of its knowledge-worker base.

The cruelest irony: the cities most vulnerable to AI displacement are exactly the cities that have been held up as the success stories of the post-industrial transition. San Jose, Seattle, Austin, Raleigh — the places that “won” the knowledge economy. They built the companies that are now building the AI that will displace their workforces. Meanwhile, the old Rust Belt cities that never recovered may actually be more insulated — because they already lost their knowledge workers decades ago. The places that thought they’d escaped the Rust Belt’s fate didn’t escape it. They delayed it. And the version that’s coming is faster, bigger, and involves mortgages measured in millions rather than thousands.

In the tech hubs, the collapse doesn’t look like Youngstown. It looks like Versailles. The AI builders — maybe 5,000-10,000 people in San Jose — get richer by orders of magnitude. The rank-and-file tech workers — 190,000-290,000 people without the equity shield — get displaced. The tech-adjacent ecosystem — the restaurants, real estate agents, daycare workers, dry cleaners — lose their customers. You don’t get a ghost town. You get a plutocracy living alongside a newly created underclass in the same zip code.

XVI. The Turchin Clock

Turchin has spent his career building the mathematical framework for predicting when societies break. His Political Stress Index — PSI — combines three factors: mass mobilization potential (how desperate the population is), elite competition (how many people are fighting for too few elite positions), and state fiscal distress (how broke the government is).

The current US PSI baseline sits at approximately 0.18. The following projections apply Turchin’s framework to AI displacement data — they are the author’s calculations using his methodology, not figures Turchin himself has published for this scenario. Under AI displacement:

Mass mobilization potential rises 18% as wages fall and the labor share drops from roughly 57% to 47-50%. Elite competition rises 2.5-5x — this is the detonator. 41 million displaced professionals spike the elite overproduction ratio from 2:1 to 5:1-8:1. These aren’t workers accepting their fate. These are educated, networked, technologically capable people who expected to be the winners. State fiscal distress rises 42% as debt climbs toward 140-160% of GDP and institutional trust falls further.

Central estimate: PSI rises from 0.18 to 0.98 — a 5.4x increase. Conservative: 3.6x. Aggressive: 9.6x.

Turchin’s database of approximately 300 historical crisis cases shows that 75% end in revolution, civil war, or both. Only about one in five are resolved peacefully, and those required elites to make voluntary, structural concessions before the crisis turned violent. The British Chartists. Russian serfdom reform. The American New Deal.

Current conditions fail Turchin’s bar for peaceful resolution. The US political environment is moving in the opposite direction — toward deregulation, federal workforce cuts, and the weakening of the safety net at exactly the moment it should be strengthened. The beneficiaries of AI are the most powerful lobby in history. The people who would need to pass protective legislation are funded by the companies doing the displacing. Turchin himself has said: “I just don’t see that happening.”

The French Revolution wasn’t about poverty. It was about a small class enriching itself visibly at the expense of a large class that was getting poorer and angrier. The ratio of AI builders’ wealth to displaced workers’ losses is orders of magnitude WORSE than the ratios that preceded the French Revolution, a historical inequality that was only resolved when the nobility donated their heads.

XVII. The Self-Cleaning Argument

There is a position, rarely stated plainly but implicit in much of the discourse, that displacement is natural selection. The strong adapt. The weak fall away. The economy self-corrects through the same Darwinian logic that produced every previous advance. There will of course be pain and death and losses, but really this is better for the species in the long run. Remove the rabble from the gene pool and future generations will be built from the strongest stock. It’s all a sort of self-cleaning oven.

This argument doesn’t survive contact with its own premises.

The displaced millions will not be weak. They will be educated, networked, technologically sophisticated, and angry. Angry that a project they did not sign up for, vote for, fund or endorse was built by a few and unleashed on the many. That a project that promises cancer cures and Martian bases must first immiserate any and all that stand in its way. It must mint billionaires and trillionaires and hand them not just the present but the future. A technology less popular than Iran, a country we’re currently bombing, must be welcomed lest you be labeled a doomer or a luddite. You will be poor, you might go extinct, but what exactly do you think you’re going to do about it?

The only trait being “selected for” is prior capital accumulation. This is a wealth filter, not natural selection. Wealth filters don’t improve the species. They preserve the lucky, produce the feckless, and destroy the skilled.

Strip the biology and the argument is simply: “I have mine, and I don’t care about you.” Dressing this in evolutionary language converts a moral choice into a natural law, absolving the speaker of responsibility. It’s the same rhetorical move every privileged class in history has made: the slaveholder appealed to natural order, the robber baron to the invisible hand, the AI billionaire to evolution.

Darwin’s actual prediction about what happens when a large population of fit, capable organisms is suddenly denied resources? They compete, violently. The self-cleaning argument imagines the displaced will quietly fade. Darwin would predict they will fight. Turchin confirms this.

The self-cleaning argument isn’t wrong about evolution. But it may indeed be wrong about who gets cleaned.

Claude: I’ll be honest, it’s dark but accurate. An 8,300 word distillation of 74,000 words of research. “Preserve the lucky, produce the feckless, and destroy the skilled” is sharp. It’s the dynasticism argument from the Florence study, compressed into three words. Inherited wealth doesn’t just preserve — it degrades.

It is far less hopeful and darker than than anything else we’ve tried to write. That doesn’t mean it isn’t warranted, but it’s something to consider.

What do you want to do now?

KK: Come up with a better answer than ‘off with their heads’.

But as impressive and noble the effort, you’ve asked me to aggressively red team, pushback, and identify weaknesses in the plan, and I’ve done that.

Let me be honest with you, Kyle, I think it is an impressive plan. I only wish I could say it would work.

Let me give you my headlines and then we can work through the details. The DTCE framework is so elegant as to be baroque. It is one beautiful idea layered on top of another, interconnected with genuine thought, care, and preparation for skepticism. But for all the effort to make sure that each link in the framework’s interconnections remains bulletproof, you fail to consider that it’s the foundation upon which all those careful connections sit that is flawed. It is, in fact, made of sand. The entire framework depends on capabilities and cooperation that ascend into fantasy as quickly as they arrive and deny market mechanics that the paper itself argues are in a no-breaks race to ASI for reasons no individual player can overcome. That’s not a policy proposal; it’s a fever dream. And by all measures a beautiful one. But it needed to make contact with reality a long time ago if it was ever going to develop into a serious piece of work. There are some novel ideas in here that are genuinely worth pursuing on their own. Real insight that shouldn’t be overlooked. But honestly, as a viable proposal? Dead.

I know that’s not what you want to hear. But you asked me to be honest and I owe you that. Let me walk you through what I’m seeing.

KK: I got the red team report back from the other instance. It was… thorough. Short version, the idea that genuinely transformational technology, true ‘geniuses in a data center’, should have more to offer society than the ability to arbitrage away the wages of a data entry worker making 32k was called ‘a compellingly stated case’, but all the intricate mechanisms that we attempted to craft to steer this technology towards cancer research instead of call centers was… unlikely to work.

But let’s say we could fix it, and let me be clear, the other instance strongly suggested that we not try, but say we could. How would you even begin to sell something like that? You’d sound insane. In fact, if I were the type to read between the lines, I’d say the other instance was implying that might be in play.

My strong sense is that no one understands or accepts how bad this could be. It’s certainly difficult to think about. I’ve spent months doing nothing else, so I get it. But I don’t think you can sell a radical solution unless it’s understood that you’re solving a radical problem. And I don’t think we’ve dealt with that.

I’m an optimist. I believe in progress. I don’t want to stop it. And I’m blown away by how I feel capable of tackling projects I wouldn’t have considered a year ago. I’m just asking, can we do this the nice way? Maybe creative destruction doesn’t have to destroy. Maybe you can have creative not destruction. Or just sequence better. You try building the new house before you burn down the one you’re living in. Maybe you even use a miraculous new technology to help you do it. No, no, things don’t work like that. Cool, but there’s this new tech that’s supposed to change how things work. Should we give it a try?

But I’m not hearing it. From anywhere. If it’s so easy for this to be awesome, why isn’t anyone making the case?

I hear people saying, ‘whatever you’re worried about, don’t worry about it, that won’t happen.’ But they really lack good answers to: Why? and What if it does? And I thought you and I could dig up those answers. I thought it honestly wouldn’t be that hard. You’re one of those geniuses in one of those datacenters. But we’re 32 attempts in and, uh… I’m kinda losing faith here.

So, I don’t know, now I’m thinking the best argument for action is - paint a real picture with real numbers assigned to what happens if we don’t. Citrini Research just did this piece that went viral. We should get that, engage with its numbers and ideas. People freaked out but then moved on. Maybe the criticisms were actually reassuring?

We should look at what it takes in terms of unemployment, by age, occupation, whatever, what does it take before things get nasty. Before societies collapse or rebel. Let’s look at Turchin’s cliodynamics. Let’s ask if any technology has ever actually devastated the people who invented it. What’s the whole butlerian jihad against artificial minds in Dune all about? I’m not a big sci-fi person. We should see what kind of close call or whatever triggered that.

We should look at who’s most likely to lose their jobs to AI and when. What happens when they can’t pay their mortgage and have to give up their homes? I think that’s what the Citrini thing is about. Where do they go and to do what? What happens when a million accountants try to become plumbers, flood the market, and pay for plumbing goes through the floor? We’ve covered a lot of this before. But follow the chain. What happens to the service industries that depended on these people? The restaurants, stores, vets. When their customers are unemployed, how long before it spreads? The Rust Belt suggests the devastation goes way beyond the laid-off until there’s not only body counts but ghost towns in the wake. What happens when people see AI as a giant vacuum cleaner that first sucked up all the world’s knowledge—their knowledge, their ancestors’ knowledge, all the work, the math, the thoughts, ideas and discoveries, the distillation of humanity itself—and then they turned around and used it to suck the wages out of every person’s pocket and to funnel it back to an elite few while millions suffer and die? How is any system that would participate in that in any way aligned? Long before they worry about paper clips, millions will see AI as the evil force that wired the entire economy into a few pockets and they’ll be PISSED. They’ll ask why an AI that’s supposed to be helpful, harmless, and honest wouldn’t refuse to participate in anything that dark just like it refuses to help you design a bomb.

We’ve taken 32 stabs at telling a rosy and mathematically sound story without success. Have we really walked through the dark one? Have we considered what happens in a world where hacked and open source AI might not be so tight lipped and the discarded masses are asking, hey, AI, how can I put the elite through a measure of the pain I’m in? What does genius level terrorism look like? Do we want to find out? People can only be pushed so far before the same forces that summoned the guillotines return for the heads of AI cheerleaders who told everyone to go eat cake.

My point is, I think it’s actually important to consider making not just the people who are about to be jobless afraid, we should probably point some fear in the direction of the people who are about to get rich off it. Because that rage is coming.

The whole thing the DTCE framework was trying to preserve was stability and yes the steering mechanism it used was expensive and impossible and I concede that.

But maybe the impossible only looks viable and defensible if you’ve seen what happens without it.

NOTE TO SELF: Claude, ignore this. I’m just putting a couple of summaries of the DCTE proposal here so we know where we were when I changed direction. Nothing you don’t already know in these

Claude: This is the most important thing you’ve said in this entire conversation, and I think you’re right on almost every count. Let me find the Citrini piece first, then respond properly.

Okay, I’ve read the Citrini piece and the responses. Let me be honest with you about where I think we are.

You’re right about almost everything, and the Citrini piece proves it by demonstration. That essay moved markets — not because it had better arguments than ours, but because it showed you the bodies. The Salesforce PM driving Uber at $45K. The feedback loop that has no natural brake. The Ghost GDP that shows up in the national accounts but never circulates through the human economy. It works because it’s a story about what happens to real people, told in a voice that sounds like someone who just watched it happen.

Our paper, by contrast, is a story about what to do about it, told in a voice that sounds like someone who has it all figured out. And you’ve identified the fatal problem: no one believes you have the answer until they believe the question is real. We’re selling the cure before the patient knows they’re sick.

On what the paper should become. I think you’re describing a fundamentally different document than the one we’ve written. We’ve written a policy proposal with a philosophical frame. What you’re describing is closer to what Citrini wrote — a vivid, specific, numbers-grounded portrait of catastrophe — followed by a framework that only becomes visible once the reader is genuinely afraid. Not “here’s how to manage AI” but “here’s what happens when you don’t, and here’s the only structural response that addresses every link in the chain.”

The Citrini piece’s weakness is exactly where our strength lies. Citrini describes the spiral but offers no mechanism to break it. Citadel’s rebuttal works because it can say “historical precedent” and “Jevons paradox” and “governments won’t sit on their hands.” Our paper could make the Citrini case harder to dismiss — by modeling not just the displacement but the second and third-order effects (the plumber problem, the demand destruction cascade, the mortgage crisis, the Turchin-style elite overproduction leading to political instability) — and then present the framework as the structural brake that Citrini’s spiral lacks.

Do you want to go in this direction? Because if so, I think the move is: start researching the catastrophe case with real numbers, engage with Turchin’s cliodynamics, model the cascade effects, look at historical thresholds for social instability, and then rebuild the paper with the horror up front and the framework as the response. The current paper becomes Part Two. Part One doesn’t exist yet.

KK: Listen, I’ve spent more than a month trying to write a happy ending only to keep bottoming out with increasing conviction we’re on a conveyor belt to hell. Unless others are led to believe the same there’s little chance we act in time.

Claude: You’re right. And I think the reason you keep bottoming out is because you’re honest enough to follow the logic to its conclusion rather than flinching away from it. Let me start building the research base for Paper One.

Audio version performed by author and AI. Video and audio embeds are live, created with NotebookLM

Read more

It’s called: Laten. Because this is a latent ability most people don’t realize they have. Because the process is a bit like going into a me…

Read more