It’s obvious that AI is already superb at writing code and at finding or exploiting vulnerabilities, and it’s only getting better. The question is what this means for cybersecurity: does everything just get hacked and we’re doomed, or do we also use it for defense and end up balancing out?

My view is the boring one (latter), for two main reasons:

(All of this assumes AI keeps getting democratized in an open ecosystem, which seems almost inevitable now. The biggest risk will always be centralized control of such a powerful technology.)

• AI is still a tool. The ultimate tool maybe, but still something that responds to prompts, and that won’t change until we get to AGI and it can actually disobey us (which I believe is still very far away and won’t come from LLMs, but that’s another topic). For now, AI is simply something humans use to augment their abilities, just like many powerful tools we’ve invented before.

• AI is not a weapon. A lot of people compare AI to stuff like nuclear weapons, like if we’re birthing such a dangerous tool that unlike nukes, it is going to be available to everyone, so it only takes a few bad guys to hack everything or create bioweapons to kill us all, that couldn’t be further from the truth. Weapons’ only use case is destruction. AI can be destructive, but can also be just as good at defending, and it’ll also be used for many other things that’ll change the world for the better.

Because it is a promptable tool that is excellent at both attacking and defending, it’ll largely end up balancing itself out and we’ll end up in roughly the same place we are now, with a few exceptions on both sides.

By ‘roughly the same place’ I mean: we have always had competent and incompetent builders and competent and incompetent hackers. There have always been situations where the attackers were better than the defenders and won and situations where the opposite happened. That will keep happening, AI is just a tool that’ll supercharge their abilities. This applies to both existing code that can be updated reasonably quickly and to new code being written today.

The exception for the bad side and probably the biggest risk right now is in existing code that is hard or impossible to update. The Coldcard incident is a great example: attackers found a bug in the way seed phrases were created and stole over $100M. Even if the good guys had found it first, they would have needed to make it public so users could move funds, and with that being public, attackers would still have had a window to exploit it. These situations are the riskiest for users and companies as even if the good guys find the vulnerabilities, they can’t patch them with code, the only fix is to migrate every user or data using that specific system and coordinating such a big migration silently is almost impossible.

The exception for the good side is that we will arguably be in a better place for small teams. Before AI, a small team would face a larger group of hackers and be outnumbered, now they can run swarms of agents that audit their code constantly, thus making the playing field more equal.

Subscribe now

The other major risk we face is just incompetence. If a company with updatable code doesn’t even bother to care about security (many such cases) or using AI to audit itself, then AI just helps the attackers which can now hit more targets in the same amount of time. Those companies are worse off than before.

On the other side: the cost of continuous AI auditing is low and the reward is extremely high. Builders have a first-mover advantage by being able to use these tools to audit their code before anyone can see it, and you should definitely throw agent swarms at your code before launch and on an ongoing basis afterward. Daily audits should become routine and the way for good guys to stay ahead.

As a user the best thing you can do is to diversify your risk as much as possible and remove single points of failure. Crypto is one of the most dangerous spaces right now because of its decentralized, open nature and the inability to just roll it back in the event of something bad happening (although I think most of it will do just fine and end up as the most anti-fragile system ever).

Mitigating that risk can mean using multisigs that diversify your risk between multiple wallet providers (selflessly shilling my own company: Oasis Vault, though there are other great options too), or split stablecoins deposits across multiple protocols you trust (for example Morpho and Aave) instead of using just one.

Elsewhere I think you just need to assume everything online will get leaked and be as aware as possible of which data you share online.

Also, don’t get hypnotized by marketing words like “air-gapped” or “open source.” Both properties can be valuable, but neither magically makes code unexploitable. Bad code in an air-gapped device still gets exploited (as we just saw with the Coldcard incident) and open-source just accelerates the finding of a vulnerability, hopefully from the good guys, but you can’t be sure.

What it’s true is that open-source software that doesn’t get hacked for a period of time is probably the most secure one out there, open-source lindy-ness is the best property when it comes to cybersecurity.

AI supercharges both sides, the overall equilibrium probably stays intact apart from the few exceptions that I mentioned before and that are worth being aware of. Everything else is just an acceleration of the same cat-and-mouse game we’ve always played, only now both the cats and the mice have much better tools. As long as we avoid the centralization of power we’ll do just fine.

Subscribe now

At some point I just had to write about it directly.

Optionality, at its simplest, is having genuine options open and the ability to take advantage of them. We’re fallible, we can’t predict complex systems perfectly, and we’re inevitably going to make mistakes. And when that happens, we need to be able to change course and self-correct.

Progress (individual or systemic) is easier when we have optionality. Every good thing that’s ever been built or discovered came from someone having the optionality to try something new, get feedback, and adjust. So it’s not about staying put with open options, you need to be moving. Once you do, it compounds: more available options mean more attempts, more feedback, and faster learning, which accelerates progress and creates options you couldn’t see before.

At the personal level, I think of this as “strong opinions loosely held.” People sometimes see that as arrogance. In my opinion is one of the most humble ways one can behave, by accepting that:

• Things are not perfect and can be improved, so you need to keep moving and trying new stuff out.

• You are going to make mistakes, so you need to keep an open mind to reverse when that inevitably happens.

This is not against decisiveness or committing to stuff. Decisive action is usually better than endless hedging, and commitments are valuable and deliberately make other options harder. But it wouldn’t be a commitment if there’s no way out of it; you have to keep choosing to do it, and for that to be true you need to have the option to exit.

That exit option matters more than people admit. People are happier when they perceive open options and depressed when they don’t see a way out of something.

Subscribe now

At the systems level it means being as horizontal and open as possible, with general rules that don’t push one specific kind of thinking or behavior. Different problems require different types of intelligence, so progress comes from letting diverse people live according to their interests and run their own experiments while keeping the ability to self-correct.

Periods of suffering and lack of progress have historically been tied to a decrease in optionality. This usually happens when a group of people treat a complex system as if it were simple, they think they can predict the future and deliver utopia, so they force society into a closed system where disagreement is punished and people can’t try new things because supposedly they already have the best answer. The answer is obviously incorrect, but course-correction becomes so hard that it’s followed by a long period of suffering.

You can still experiment with extremely rigid systems, as long as they have an exit option, it’s ok. Imagine the amount of suffering that would have been saved if people were able to try a system like communism, realize it doesn’t work and then exit fast. The problem is never trying new kinds of systems or making mistakes, the problem is removing optionality from participants so they can’t self-correct.

So in my opinion, until I change it:

• capitalism > socialism

• free speech > censorship

• crypto > central banks

• open-source ai > closed ai

• decentralization > authoritarianism

This is why I keep returning to it. Almost everything I end up caring about shares a common denominator: preserving or increasing optionality.

Subscribe now

On the value accrual side, raw intelligence will probably end up a commodity that’ll keep getting better and cheaper as time goes on. Inside the AI industry, most value will sit in hardware (GPUs, memory, etc.), in serving that intelligence to users (inference, cloud infrastructure, routing), and probably specialized models trained on proprietary data.

The other two: how companies compete and what builders should optimize for, are even more important. And both come from the way an open ecosystem actually works. This specific ecosystem is defined by four main attributes:

• Open weights models: Anyone can run them privately and uncensored, and can also fine-tune them. They may not always run on a laptop, but any company can host them, adapt them to their own values (private, uncensored, custom rules), and sell inference. This makes models easily swappable depending on the specific need.

• Open source harness: Harnesses of all types (Codex, Grok Build, Hermes, Open Code, etc.) are already open source. Anyone can build on them without fear of access being revoked. In this scenario, closing your code only makes your product less composable and worse.

• Uncensored agents: Because they run on open models and harnesses, there are no gatekeepers on what an agent is allowed to do or say. Not just which tools it can use, but access to the internet, the computer, private keys, or anything else the user wants. Agents are free to choose whatever is actually best.

• MD files: Memory and context live in the most primitive possible format. This makes switching products as easy as copy-pasting a file.

The combination of these attributes creates a fairer, more meritocratic version of what we saw in recent years with the web2 and social media era. It comes down to two main advantages:

Ability to switch: Infrastructure being open and the valuable stuff for users being MD files that they can copy-paste anywhere gives users the ability to switch and swap services in an instant. This forces companies to behave well as they can’t abuse their power if the user can just turn around and move to a competitor.

No king-making by leading companies: In an open tool, integration is permissionless. Leading companies can’t king-make by letting some companies integrate with them and not letting others do the same, or by revoking access to anyone building something they want for themselves. Builders can build on top of products knowing they won’t lose access in the future, which keeps the ecosystem composable.

The opposite is true for social platforms. Those companies own user data and audiences. They have strong network effects: a social product with ten users is useless, but with a billion it becomes valuable because everyone you know is there. Those network effects created long-term moats and gave them enormous power once they reached scale.

AI is different. It’s a tool, not a network. Your experience with an agent does not depend on everyone else using the same one. You can use GPT on Monday and Kimi on Tuesday without losing anything of value. The inability to create network effects is what will make the AI ecosystem look similar to the early internet or crypto: open, decentralized and competitive on product merit alone.

The companies that treat AI as core infrastructure, and the builders who lean into openness instead of fighting it, will pull ahead of walled-garden approaches and slower incumbents. Look at how even Facebook and Google, despite billions invested, have struggled to compete with new and faster challengers.

And just like the open internet, the AI ecosystem will start skeuomorphic (a term Apple made famous in the 2000s, for when a new technology imitates the form of whatever came before), like how early newspaper websites looked like printed papers and updated content just once a week. AI is doing the same thing right now by trying to replace one human customer service agent for one bot doing the exact same job.

Over time it will go native and unlock new use cases that were hard to imagine beforehand. Amazon’s everything store with reviews and personalized recommendations was unthinkable in the pre-internet world. The native use cases for AI are still coming, and the biggest ones are probably things we can’t name yet.

Even if we don’t know exactly which new industries and native use cases it will unlock, we already know one thing for certain: it’s going to be open. And that should already change the way business owners think and act about what they’re building:

• Build products that enable model swapping from day one. Never depend on a single model staying the best.

• Make it easy for anyone to integrate with your product: treat the ecosystem like lego bricks, the more composable your product is, the faster it improves.

• Invest in branding: It is still underestimated in tech, but becomes extremely important in high-competition environments.

• Be the first: When everyone can build quickly, being first to a strong idea matters a lot.

• Market to agents: Uncensored agents will have the freedom to choose which tools they use, so be discoverable where agents look for tools.

• Build tools that uncensored agents will prefer: for example, a censored agent might be pushed toward a credit card because of a partnership, an uncensored agent will often choose crypto to save fees and time.

• Play long-term games: when switching is so easy, short-term hacks will not hold, you’ll have to keep improving the product to retain your users.

An open ecosystem doesn’t kill competition, it just makes it honest. When users and agents can leave in a second, the only moat left is being better. The best, most composable product wins. And that’s how you actually get better and cheaper products for everyone.

Time is accelerating, if you measure it by the number of events that happen every second, it’s passing faster than ever. A second hasn’t gotten shorter but the number of events happening per second is growing exponentially. A year now contains more change than a decade used to.

The world was already too complex to predict specific futures. The explosion in events is making it even harder. Every new discovery or invention expands the adjacent possible and creates possibilities that did not exist the day before. We can sometimes predict general directions (problems remain solvable, change is inevitable), but precise prediction has become almost impossible.

And somehow we still have a group of people that insists on regulating the rest of us according to the future they see.

We’ve made this same mistake many times in the past, entire political systems are based on the belief that society can be predicted and controlled. And time and time again we’ve seen the opposite: progress comes from the freedom to think, combining ideas, and running open systems that can self-correct, instead of closed systems trying to prevent mistakes. Aiming for utopia has always left us poorer and less free.

Closed systems require two things from the central controller: good intentions and the ability to predict and optimize the future. And because they are built around a specific prediction, they are inherently static and inflexible. When reality diverges from the plan (it always does), they can’t adapt.

Open systems work better because they don’t require anyone to be a prophet. They are dynamic. They assume from the start that we are fallible, that problems will appear, and that most of our current ideas are incomplete or wrong. Instead of trying to prevent every mistake, they maximize the ability to detect and correct errors.

Bad things still happen in both, but open systems are built to adapt. They treat problems as solvable rather than as justifications for more control. They handle uncertainty better.

Subscribe now

We’ve never had more uncertainty than right now. AI is the ultimate accelerator, it’ll help us generate new knowledge and new tools at a rate never seen before. It will also create new problems, that’s inevitable. And we can either design these new systems to be as open and self-correcting as possible, or bet everything on our ability to predict and control what comes next.

On the closed side we have governments pushing for KYC to use the internet and banning companies from offering their services to the world. Companies claiming that AI should be closed and controlled by them because, of course, they are the only ones capable of building God.

All of it pushed by doomers who believe they are smart enough to predict future catastrophes: AI turning us into paperclips (via a dumb goal in the most intelligent system ever), average people creating bioweapons in their garages (while somehow not helping billions defend against them), or a permanent underclass (thanks to the most rapidly democratized technology in history).

On the open side we have the historical record: capitalism outperformed communism, democracy outperformed authoritarianism, the scientific method , the open internet. In politics, economy, science, and technology, open systems have consistently beaten closed ones.

They are messier and bring problems, and that’s the reason why people keep trying to build perfect, problem-free systems from the top down. But problems are solvable and open systems are great at that. Utopian projects are at best naive dreams, and at worst propaganda for more control.

We already have the tools required to push our systems in the open direction: open-weight models that anyone can run privately and uncensored on open-source stacks, crypto as the infrastructure to have decentralized, trustless systems, and the internet as both the base layer for all human knowledge and the clearest proof that a messy open protocol can change the world for the better.

The future is unimaginable as Stewart Brandt says, and that is precisely why we should refuse systems that require someone to imagine it for us, and push for systems that maximize freedom and optionality for all.

Subscribe now

When models started getting really good at coding a few months ago, I, like thousands of others, got obsessed and started vibe-coding every idea I had like my life depended on it. Having access to a tool that could build whatever you described was an amazing breakthrough that unlocked so many ideas in people’s minds.

From January to April I vibe-coded and used regularly:

• My personal email client

• My personal agent harness (forked Openclaw and gave it some privacy/security enhancements + a personal desktop app to access it)

• A phone widget so my agent could control my home screen and push relevant info to

• Plus ~10 other personal and work tools/projects

So naturally, all sorts of theories about future consequences started to form, one of the most popular being that SaaS was dead. I also believed it. Who’s going to pay thousands of dollars a month to different softwares that you can now build yourself?

And then some time passed. And the AI psychosis went away. And I started to normalize having the ability to build whatever, whenever. And right at that moment… my tools started needing some sort of maintenance.

I knew about this need but somehow ignored it. I’ve been building digital products for years, and “we need to maintain this” and “we have technical debt” are two of the most common sentences you hear from engineers.

Once you build a product, you can’t stop maintaining it, not just because you always think of new ideas you want to implement, but because you inevitably have dependencies you used to build it. Libraries, APIs, all of it requires you to maintain the code and keep it up to date.

Maintenance takes a lot of time, and even if you’re not writing the code, just the job of prompting plus testing can take hours a week. Once the excitement dies off, it becomes a burden, just like how you can’t wait to learn to drive, then eventually just want to be driven; or how as a kid you wish for an agenda to manage, then grow up wanting to free it. Same thing with vibe-coded projects: once you get used to them, you just want them to do their job without you having to intervene.

Luckily, just like we have Uber if we don’t want to drive, we have this thing called SaaS, software-as-a-service, but it’s basically maintenance-as-a-service. You’re paying someone else to keep the tool alive, either by maintaining it or by spending their time thinking up and building the features you need. And because software is scalable, they sell the same service to millions for a small marginal cost. Win-win.

So yes, I’m back to my usual tools. I did get to keep a few custom ones for work that we decided were unique enough to be worth maintaining, and I’m still using my agent daily (Hermes now, plus some custom plugins on top). Agents keep improving (both the harness and the underlying models) and are going to be a big part of every person and business really soon. They’re actually the biggest threat to SaaS by abstracting the front-end. But they aren’t ready yet: their Achilles’ heel is still completing deterministic tasks, and that’s exactly where SaaS excels.

I do think that AIs being so good at coding will change the world in many different ways that we can’t even imagine. We still haven’t found something that’s not programmable in the universe.

And it has already changed the developer job forever, their productivity has increased by multiples. But when it comes to companies or individuals building their own tools just because they can vibe-code them, I don’t think the maintenance, the token costs, or the attention needed to do it is going to be worth it anytime soon.

So yeah, SaaS might not be dead after all.

There’s this idea of the superhuman AI, the one that gets to a certain threshold and just accelerates and improves itself forever, the one that either kills us all or makes us immortal, one AI to rule us all. We can thank science fiction writers and the doomers for that vision.

I don’t buy it.

That whole idea rests on the false assumption that intelligence equals power or success. If that were true we’d live in a world run by physicists and mathematicians. Instead, we get [insert your favorite mediocre politician here], in the real world you need a combination of skills to make things happen.

IQ itself is overrated when it comes to usefulness. It helps and it’s probably the one trait that people can’t learn. But just analyzing stuff doesn’t get it done, Kevin Kelly calls this thinkism: the belief that thinking about things will solve them. And when people worry about the superhuman AI taking over they’re making the same mistake, assuming that just because it’s AI, IQ will suddenly matter more.

What we call AI today is narrow AI: tools that are incredibly powerful at specific tasks but incapable of “taking over” anything. The superhuman AI scenario only makes sense if we’re talking about AGI: a fully autonomous system with real creativity and agency. But when we eventually get to AGI, I think we’ll realize it’s just another person with better hardware.

The part that makes an intelligence general is creativity: the ability to create new knowledge, as David Deutsch says, and that’s software, not hardware. You can’t brute-force your way to creativity with more compute.

Humans are already general intelligences with that software, the difference with AGI will be hardware: more speed, more memory. But there’s no reason to believe humans can’t also take advantage of that same hardware.

Subscribe now

So yes, the first AGI will be a genius, probably the smartest mind on the planet. But one genius, no matter how capable, is still just one person, clone it a million times and the marginal value of each copy drops toward zero, the law of comparative advantage makes sure of that, people are valuable because they are different.

We do need the high IQ physicists, mathematicians, engineers, sure, but we also need philosophers, artists, teachers, salespeople for a society to actually move forward. Different problems require different types of intelligence.

The “one AI to rule us all” scenario doesn’t make sense. Intelligence isn’t winner-take-all. Reality is way more interesting than that, what we’ll probably get instead is what we’ve always gotten with new technology: specialization and diversity. Many different AGIs and AIs working alongside humans solving problems faster than ever.

Subscribe now

I’ve been in crypto for more than a decade now, and sentiment among natives right now is the worst I’ve seen. Generally in past bear markets, the tourists would leave and the natives would stay, either building something or just questioning their life decisions in group chats and CT. This time it’s the opposite. Tourists, in the form of fintechs and institutions, are rushing to integrate stablecoins, DeFi, and tokenize everything they can put on-chain.

And on the other side, this is the first time I’m seeing crypto natives actually leave the space. One reason is AI, we haven’t had a cool new technology like that since Bitcoin and it’s only natural to lose some talent to it. But others are genuinely burnt out and disappointed and fed up with the state of space right now and what they see as crypto’s future. There’s some part of me that gets this latter group, so it feels like the perfect time to lay out where crypto actually stands and where it’s heading.

The two sides of crypto

I see crypto as divided into two big categories, which for the purpose of this article I’ll call cypherpunk and fintech:

• Cypherpunk: Decentralized, permissionless, private, and uncensorable networks that work as an exit valve from current systems, both political (by exiting fiat) and corporate (by having open protocols where users and builders don’t depend on a company’s permission to build a product or migrate their data).

• Fintech: Just better tech that helps users and businesses lower costs or work better. Cypherpunk attributes also help this use case (they make the actual product better, I’ll say more about this later), but they’re not as important as in the other category.

The narrative that crypto must be one or the other is ridiculous. Crypto can be both at the same time, they just need to be judged and valued differently. I’m more sympathetic to the cypherpunk side; that’s why I got into it and what I’m working on now. But having worked in fintech for years, I also understand the value of using crypto rails (even with centralized assets) to improve people’s lives.

Why natives are leaving

Some natives are exiting because the cypherpunk part seems to be disappearing, or at least weaker than ever. Five specific concerns come up most often:

• Institutions did come for our coins and now we’re angry about it: It was always inevitable that they’d get in, if crypto becomes the backbone of the financial system, institutions have to be part of it. But the way it happened this cycle wasn’t great. Overleveraged DATs buying crypto with debt instead of productive companies earning real revenue and allocating profits into it. Markets will probably take care of this and it’ll fix itself with time.

• TradFi and Fintech companies “stealing” the tech: The idea that we built open protocols, and now that they’re good enough, traditional companies will take them and integrate them into their systems without any cypherpunk values. This can’t really happen at scale (maybe for some specific zk use cases), because the tech fundamentally doesn’t make sense if it’s not decentralized. We’ll get to that later.

• No decentralized mainstream money: BTC is probably not going to be used as money. Stablecoins will fill that role for the next few years. They have defects: not fixed supply, not decentralized, not permissionless. But they’re a big improvement over the status quo and generally a net positive. More on that later, also.

• Market nihilism: Not what you want to see as a builder. Two reasons: regulation made it so real products couldn’t launch a token fairly and pushed speculators to meme coins, and almost all important primitives were built before 2022. The invention of new primitives has historically been the best time to invest in crypto (ICOs, Defi, NFTs), and we only had one big one this cycle: decentralized perps via Hyperliquid, and it did pretty well.

• DeFi is riskier than TradFi because of AI: There will be some exploits, but DeFi will survive, use AI defensively and end up as the most antifragile system in the world.

There are some valid claims for all of these and we probably won’t get a cypherpunk utopia, but the overarching claim that the cypherpunk part in crypto is dying is overblown, here’s why:

Cypherpunk values come down to optionality

Cypherpunk values, at their core, come down to optionality. The freedom to always be able to choose what you want to do with your life. The freedom to exit.

But optionality isn’t a given. To have real optionality, you need infrastructure that most people take for granted until they need it. You need privacy. You need decentralized networks. You need money that can’t be frozen or inflated. And you need protocols that don’t depend on corporations or governments to keep running.

This is why cypherpunks fight so hard for things like privacy and decentralization. They don’t want to force everyone to use them, they just want to build the tools because without them, optionality collapses. I remember Zooko saying once that he has no more personal secrets while at the same time building privacy tools for a living, or libertarian Riva Tez saying the endgame is radical optionality, and that she wants to build tools that even communists can use to experiment. Those always stuck with me as the perfect examples of cypherpunk values at work.

Optionality at its core means always having the ability to exit or switch. This is true in the political space, as Balaji writes about: crypto serves as an alternative to fiat, a check on the political system. And it’s true in the business space, as Chris Dixon talks about: open and permissionless protocols let users and entrepreneurs build and switch whenever and wherever they want, without depending on the permission of centralized corporations to do it.

These tools are defensive technologies, and all we need is for them to exist as an option. In a hypothetical world where governments and companies act perfectly, these tools would be worth almost nothing. The value is in having the option to exit, not in the specific price. I don’t think that optionality has changed or weakened in the last few years. Actually, it has strengthened along with the reasons for these tools to exist. Self-custody tools are more accessible than they’ve ever been, privacy tech is meaningfully better than five years ago, transacting on decentralized infrastructure is cheap enough that it’s a real option for most people, all while centralized authorities keep giving us more and more reasons to build it.

The fintech explosion

Optionality doesn’t just mean the freedom to exit, it also means the freedom to build on top of it in whatever way makes sense. Permissionless networks don’t discriminate between use cases. The same infrastructure that lets someone exit fiat also lets a company build a more efficient payment rail. The same composability that lets you self-custody your wealth also lets a fintech app plug into DeFi without asking anyone’s permission. This was always the point. Building open, permissionless networks doesn’t just enable the cypherpunk use case, it enables the fintech one too, it was always inevitable. And right now, that specific space is exploding.

Every finance company in the world is thinking about or already integrating crypto rails into their products in the form of stablecoins, tokenization, or DeFi. And the value for them has nothing to do with cypherpunk values. All they want is to grow their business by either lowering their costs or improving their user experience.

Crypto infrastructure is now ready to handle millions of transactions with high speed and low fees. That enables all kinds of use cases by removing intermediaries that cause a lot of problems for regular users and new fintech startups, by charging high fees or just being too bureaucratic to give access. The decentralized and permissionless nature of DeFi combined with the stability of stablecoins makes it so international payments can be almost 95% cheaper. Companies can now offer users access to dollars and yield without having to ask for permission. And hopefully stablecoins get adopted as a payment method and businesses around the world start saving the 2-4% they pay to process credit/debit card transactions by using crypto as the settlement layer.

Fintech needs decentralized rails

So where does this leave us? Both the cypherpunk and fintech sides of crypto keep growing. They serve different needs and should be valued differently, but they both benefit from the same underlying infrastructure. And fortunately, that infrastructure holds cypherpunk values.

Fintech products are going to get built on decentralized rails, not because these corporations care about decentralization, but because all these systems are better when they are interoperable between each other and finance companies won’t build on their competition’s blockchain, they’ll want credible neutrality. The moment one institution owns the infrastructure, every other institution has to decide whether to trust them for final settlement. They won’t. They’ll demand neutral ground, and trustless settlement between parties that don’t trust each other requires decentralization by definition.

Decentralization and credible neutrality are attributes that tend to emerge with time in crypto and that’s because of the culture those builders have, you could assume Ethereum was going to decentralize itself after the dao fork because of Vitalik’s and other eth builders culture, same with Solana via Mert and Toly, probably same will happen with Hyperliquid, but i don’t think you can assume that’ll happen with corporate blockchains. It’s hard to make decisions in favor of decentralization in big centralized companies like Circle or Stripe.

I’ll take a small tangent here and give my opinion about the valuation debate that’s been going on for the last few years. The cypherpunk and the fintech sides should be valued differently. You can’t have just one valuation model for completely different fundamental use cases.

On the fintech side I agree with people like Yano and the blockworks team that say we need to value these tokens as we value companies, based on clear business metrics. And that’s because the whole value of those products/companies depends on it generating growth and revenue, a centralized app with low activity or growth has zero value.

On the other side as I explained before, a cypherpunk product is a defensive product, an insurance of some kind, what you want is to have the option to use it if you ever need it, so even if it doesn’t have a lot of activity for a period of time it still retains some value, and you can calculate that value in two different ways:

• For SOV assets like BTC or Zcash is the necessity of that exit/insurance for the market at any given time, so bitcoin is more valuable if governments are misbehaving and losing the trust of its citizens.

• For smart contract blockchains like Ethereum or Solana, it comes mainly from the total value they are securing, so if Ethereum gets even more adoption and is securing trillions of dollars in stablecoins and other tokenized stuff, the valuation of ETH the asset, even if it’s not generating that much revenue or activity, should increase to at least the point where attacking it is not a reasonable action for a bad actor.

How this plays out in practice

With that in mind, let’s look at how this plays out in practice. Here’s how I think about each of the major crypto categories, where the cypherpunk value lives, and where the fintech value is being created.

Store of value:

By store of value I mean crypto assets that are primarily thought of as currency or digital gold as their only objective. Bitcoin and Zcash being the main ones. Limited supply, decentralized and permissionless network, a hedge against governments printing money and destroying your wealth or a hedge against a dystopian surveillance future.

Bitcoin is still the biggest one here and probably stays like that for decades. Biggest network effects and biggest brand as the limited decentralized currency. But it has a few short-term problems to work through.

Saylor being the most obvious. I don’t think the math on Strategy holds long-term. You can’t raise infinite money to keep improving BTC per share forever. Eventually you reach for expensive and short term debt. When those products fail and they lose the ability/trust to keep raising money and by consequence the ability to increase BTC per share, they become irrelevant. Why take the risk of holding MSTR when you can hold real Bitcoin or direct exposure to it via the ETF? Whether they sell or hold the Bitcoin doesn’t matter as much as removing that centralized dependence and market overhang, that will be bullish for Bitcoin long-term.

Then there’s quantum and the security budget, both real problems and Bitcoin maxis are probably the worst group of people to coordinate a solution on. They’ll get there, but it’ll be slow and painful. At the end of the day, if there’s a network sufficiently strong and anti-fragile to go through these problems, that’s Bitcoin. It’ll be rocky for some time, but Bitcoin will be just fine in the end.

Zcash is the other store of value I think about seriously. Privacy is going to matter more as surveillance states mature and AI makes mass monitoring trivially cheap. The recent bug situation showed the downside of being this good at privacy. If there’s an exploit, you can’t verify whether it was used, but the team is probably the best in the space to handle exactly that problem. If they can formally verify supply integrity and keep building on it, Zcash has something Bitcoin doesn’t: real privacy. Bitcoin will survive its rocky road. Zcash, with a different use case has a genuine growth story ahead.

DeFi

DeFi is the most important use case in crypto. Having store of value assets matters a lot more when you can also borrow against it, earn yield, and trade it without asking anyone’s permission. The alternative, holding an asset you can only access through a centralized intermediary, defeats most of the point.

The current narrative that DeFi is just too risky feels overblown. The biggest protocols have had billions sitting in open source code for years, the best hackers in the world have had every incentive to exploit them, and haven’t. Yes, AI is going to stress-test these systems harder than anything before, but it cuts both ways. DeFi will use those same tools defensively, and what survives will come out the other side of that as the most battle-tested financial system ever built.

The primitives already work amazingly well. We’ve had billions of dollars in the biggest lending protocols for years without an intermediary, same for DEXes and similar protocols. The whole DeFi game now is integration: how do you get this into every person’s regular financial life?

The open source and natural composability makes it easy for anyone to integrate it. Protocols become lego bricks that any app can plug into. Because of this, crypto teams are going to have to compete with fintech companies on the consumer side, and fintech companies are experts in exactly what crypto teams lack: UX and user acquisition. That competition will only benefit users.

Stablecoins and Payments

Stablecoins are one of the best products to come out of crypto. They’re not cypherpunk, I don’t agree with Haseeb on that, but they don’t need to be to be genuinely useful. They don’t solve infinite printing or reduce reliance on government money, but they do give people everywhere access to dollars, which are still the best fiat asset by a wide margin. For someone in Argentina or Turkey or Nigeria, that matters enormously. Pair that with DeFi and you can earn yield on those dollars too, something most of the world has never had real access to. And when used on decentralized rails via self-custody wallets, they’re also a better (and more cypherpunk-aligned) product than fiat, even for Americans.

They’re also just better tech for payments: no fees, faster, programmable. The average business pays 2-4% to process a card transaction. That fee is essentially just the cost of the settlement layer, and crypto makes it negligible. For a lot of small businesses, that difference is their entire margin.

The reason stablecoins haven’t been more adopted as payments is UX. Paying with a card or through Google/Apple Pay is genuinely great, and consumers don’t have a strong incentive to switch. Those habits are hard to change. But two trends will accelerate adoption:

First, payment companies are rushing to adopt stablecoins and incentives to build these new payment rails. Infrastructure is almost there. Wallets like Privy let users hold crypto without even knowing it. The moment a large company manages to integrate with Apple or Google Pay, transacting with stablecoins in the back instead of Visa or Mastercard, and giving access to millions of customers, that solves almost the whole UX problem on the user side. On the other side, businesses are too incentivized not to push this further, they’ll offer cash back or discounts at first, then just make a habit of it.

Second, AI agents. Crypto is naturally built for agents. They don’t care about UX. They only care about effectiveness and cost.

Stablecoins are winning the money race for now, and as long as they remain on decentralized networks with self-custody as a possibility, this is a large net positive for the world.

AI

The combination of programmable money and decentralized networks leads naturally to crypto’s next major use case: AI agents and AI infrastructure.

There are 3 main use cases for the intersection of AI and crypto

The clearest one we already spoke about, is that crypto is the native money for agents. Open source, no KYC, no permission required, cheap to transact. Agents don’t need great UX, they need programmable money that works without a human in the loop. That’s crypto. Agents will end up transacting on crypto rails.

The second is decentralized inference, running open source models on infrastructure where users can actually verify what they’re running. This matters in a world where a handful of companies or governments might otherwise control AI access. Open source models are already strong and getting stronger, and governments can’t regulate what they can’t see. Providing the infrastructure to run them privately and verifiably is a big use case.

The third, training frontier models on decentralized infrastructure, I’m more skeptical of. The compute required to train these models is going to data centers via orbital satellites in space with near-unlimited solar energy. Aggregating consumer/gamers GPUs isn’t going to compete with that. It’s a cool idea, but I don’t see it working in practice.

Tokenization and Perps

The core value of tokenization is access. Even blue chip US stocks are hard to buy outside the US, and most global financial instruments are even more restricted. Tokenizing them doesn’t just solve access though. It unlocks 24/7 trading and real self-custody of the underlying asset. Being able to hold a SpaceX share, borrow against it in DeFi, or transfer it peer-to-peer without needing to sell it back through a broker first is a big improvement from what exists today.

Tokenization not only works to tokenize existing stocks or commodities, it also unlocks cool use cases for new businesses, a recent great example is Venice founded by crypto OG Erik Voorhees, they have 2 different tokens, one that serves as “equity” to the business and its tied to their revenue and the other serves as inference, granting users x amount of it a day by holding it. These kind of use cases will keep appearing as a way for businesses to get liquidity (in good and bad ways, we’re in a free market after all) and for users to get access.

For trading, perps are the more interesting product. I learned the hard way on Bitmex that I’m not a trader, but have played around with products like Hyperliquid and Trade XYZ and they have genuinely impressed me. They’re not 100% decentralized (they’ll probably trend towards it, as I said before), but decentralization is always a spectrum. Having the same product quality but with no KYC, transparent enough and with proof of reserves is a big improvement over centralized exchanges.

Prediction Markets

People have been trying to build prediction markets for 10 years now, it took this long because the infrastructure wasn’t there. Now it is, and Polymarket proved the concept by going mainstream. A lot of users don’t even realize it’s crypto, which is the best possible signal.

Most of the volume right now is pure gambling, and that’s because these platforms grew fast, valuations got big, and now they need volume to justify it. That’s completely fine. But the real breakthrough use case is surfacing expert knowledge that can’t be shared publicly for political or institutional reasons. A scientist who knows something but can’t say it in the media for career reasons can still bet on it anonymously. That information surfaces in the price. That’s genuinely valuable for the world.

It’s already becoming true that when you want to quickly check whether a news story is real, you check a prediction market. These platforms will keep growing, probably with more gambling features because that’s where the revenue is, but as long as they preserve the truth-surfacing use case alongside it, it’s a net positive.

Why I’m still here

So why am I still here?

It’s inevitable that both the cypherpunk and fintech sides of crypto keep growing. The fintech side will bring crypto to billions: cheaper payments, access to dollars, programmable money. The cypherpunk side ensures that when people need to exit, they can. The fintech side is an onramp to the cypherpunk one. Every person using a fintech app built on crypto rails is one click away from self-custody, one click away from buying Bitcoin, one click away from exiting the system entirely. The infrastructure enabling both (decentralized, permissionless networks that anyone can build on and no one can shut down) is stronger than it’s ever been.

I see the fintech adoption of decentralized rails as a big improvement over the status quo. But what keeps me here is the cypherpunk side. I don’t trust centralized authorities, and building optionality matters more than most people realize. There are many important things to get right, but three stand out to me right now:

First, ownership. Crypto is the only place in the digital world where users can have true ownership. And that, self-custody of your money, your data, your identity, is the foundation that makes the freedom to exit possible. Everything valuable about crypto is downstream of self-custody, and while I don’t think that we should force anyone to hold their own keys I do think that everyone should have that option open. There’s still a lot of work needed here to improve the UX and security of these tools and get to the point where self-custody is a viable option for everyone.

Second, privacy. It’s hard to have freedom when everyone can see what you’re doing on a transparent ledger. Having the option to transact and own assets privately is fundamental, and we haven’t really solved this in crypto yet. Zcash is a great start, but we also need privacy on smart contract blockchains for people to be able to use defi and hold stablecoins in a private way. This will get solved by adding privacy tools in wallets themselves and making it really easy for the user to do it.

Third, verification. AI is going to make it trivially cheap to fake anything. We’re entering a world where you can’t trust what you see by default. Crypto does the opposite. It makes it possible to verify things without trusting a central party. They are complementary forces for the next decade. The more AI erodes trust, the more valuable crypto’s verification layer becomes.

Crypto won’t stop being chaotic, it’s a free market and because of that a reflection on society, bad things will be built on top of it. But at least it doesn’t need bailouts, it’s a self-correcting system that keeps improving and strengthening. The main objective is to give people complete optionality to experiment however they want, without forcing anyone to participate. Achieving this gives society the tools to maintain freedom and keep progress alive.

Subscribe now

Governments and companies keep giving us reasons to support open source AI models and tools. On one side, we have governments enforcing KYC to use social media and banning companies from offering their services to the world. On the other, we have companies claiming they’re building the most dangerous technology ever while still building it because, of course, they’re the only ones capable of building god. The world would look straight out of an Orwell book if it wasn’t for (ironically) Chinese open weights models.

A world where AI is controlled by either governments or companies, where they can not only censor the models but also know what everyone is saying to them, inevitably leads to a dystopian society where a few bad actors have complete control. Open source models flip that dynamic. When weights are public, nobody can shut down your access to intelligence. You can run it locally and privately, audit it, and build on it without asking for permission.

These models are still a few months behind SOTA like GPT 5.6 or Fable, although the gap is closing quickly, as we just saw with GLM 5.2. The good news is that being a few months behind is enough to avoid an Orwellian scenario. Not only because most people can now do 90% of their work with these models (and this number keeps climbing), but also because, by existing, they serve as a kind of defensive shield against centralized control.

They achieve this by giving users the option to switch at any time, having alternatives keeps governments and companies in check, forcing them to behave in order to retain users. It’s much easier to abuse power when you have no competition.

This is the same principle that makes crypto valuable: not that it’s perfect, but that it’s an exit. Nobody has to ask permission to use it, and nobody can switch it off.

Yes, there are trade-offs. Bad things will happen in open systems, but at least they can self-correct. Closed systems don’t prevent bad things from happening, they just make them almost impossible to fix.

Subscribe now

Intelligence is (hopefully) being commoditized, users can now swap models and barely notice the difference. This has forced AI companies to shift their competitive focus from model performance to the application layer and it’s becoming clear that the winners will be those who give users the best tools to manage AI agents.

Since computing began, industries have followed one of two paths:

• User-first: web 1 (emails, domains), traditional software (Microsoft, Adobe, etc) and crypto. Open protocols and low switching costs kept companies honest. Users retained control because moving to a competitor was always an option.

• Company-first: Web 2.0, social media, search engines. Network effects are so big that companies only need to play nice long enough to achieve critical mass. Once the network effect kicked in, users became trapped.

AI is shaping up to follow the first model because of a fundamental reason: AI is a tool, not a network.

Your experience with your agent doesn’t depend on everyone else using the same tool. You can use Claude on Monday and Codex on Tuesday without losing anything of value. This eliminates network effects as a viable moat. The only path left for companies to trap users would be the Apple-style data lock-in with a closed ecosystem and painful migration processes. That’s what makes the industry’s adoption of markdown files so important.

Markdown is almost offensively simple: plain text with basic formatting syntax that any human can read and any tool can parse. Text files that have worked since 1980 and will work in 2080.

For AI agents, markdown is emerging as the standard format for context, memory, and instructions. And because markdown is just text, you can copy your entire agent configuration to a competitor in seconds.

This makes lock-in almost impossible. Your entire workflow, agent’s memory and custom preferences are all human-readable text that works everywhere. Companies must compete on quality and innovation every single day, because you’re always one copy-paste away from trying their competitor. As long as markdown stays the standard, we users stay in control.

Subscribe now

The biggest risk in AI is centralized power. As much as the EAs and rationalists want to convince everyone that an AI can convert all of us into paperclips, that we’re creating something that will treat us like ants or 7-year-old kids, that’s just thinkism, as Kevin Kelly says. The real risk is centralizing the power of such a tool to either a few companies or governments.

There are two ways we can end up with centralized control: governments or companies. On the government front, it’s not a win yet. All kinds of forces are trying to get AI regulation passed: politicians looking for more power, companies looking for regulatory capture, and even the Pope now with his latest encyclical. I think he nailed the centralization risk in companies without realizing there’s the same one when you regulate them. Actually, government centralization is worse, at least companies have the government watching over them, but who watches the watchers?

On the other front, things are looking as good as we could hope. We have all kinds of models and companies competing: SOTA models pushing the limits every other week, open weights models letting users run private and local model AI without asking for permission, American models, Chinese models, we even have a European one. Intelligence is becoming more and more like a commodity, and the app layer seems to be what’s important.

As long as models are swappable and users have the possibility to exit and switch, that risk is completely mitigated. That kind of environment leads to competition, and competition forces companies to behave well. If they misbehave, they start to lose users fast.

So now we only need to worry about the government side, and even there things are looking better than they could. Regulation happens country by country. If one country decides to lock everything down, the rest of the world still has access to unregulated models. That creates a natural incentive against over-regulation, no country wants to handicap itself while everyone else races ahead. This is already happening; China has, ironically, open sourced its models to try and damage US companies’ revenue, giving access to unrestricted models to users around the world.

All the right incentives are starting to align. The future of AI looks less like 1984 and more like the internet: messy, distributed, and impossible to control. The kind of environment that lets a thousand models bloom.

Subscribe now

Crypto is valuable because it lets people store and transact value without having to ask for permission or trust anyone in the process. That was Satoshi's big breakthrough: a decentralized ledger of ownership where participants don’t have to trust each other and can still trust the network as a whole.

That property, trustlessness, is the foundation for everything else. And nothing can be trustless without the ability to self-custody:

• Uncensorable money? Only if you hold the keys, not if an exchange does.

• Private money? Impossible by definition if someone else has custody.

• DeFi? Only if you control access, not if a company can freeze your position.

• Permissionless markets, decentralized payments… all of them only work properly because self-custody exists.

Self-custody, at its core, means that the owner of an asset isn't trusting anybody else to access or transact with it. Whoever holds the keys controls the asset, without having to ask anyone for permission.

This is already rare in the physical world: easy with cash or small stuff you can store where you live, but even the ownership of your house depends on a government deed they can revoke at any moment. It's even rarer in the digital world: you don't own the money in your bank account, and you don't own your digital identity either: your social media handles, your iCloud photos, your Google docs. You never owned any of it, you're just trusting a corporation not to take your access away.

Crypto is the only place where that actually changes. It's the only system that lets you hold value in digital form the way you hold cash in your pocket.

And the mere existence of self-custody as an option is what creates the value in crypto, even for people who never fully use it. Anyone who interacts with a crypto protocol benefits from some of its values. There's a whole spectrum:

• Buy the real coin and self-custody it → full upside, true ownership, and the freedom to exit on your own terms.

• Self-custody a stablecoin or RWA → you get freedom and anonymity a bank account will never give you, even though the asset itself is centralized.

• Hold an ETF → you get the upside of a scarce decentralized asset, but none of the ownership. An ETF of a centralized asset with "limited supply" would be worthless.

Full self-custody on one end, complete trust on the other. But the spectrum only exists because self-custody is an option in the first place.

I think the real goal of crypto is to increase freedom and optionality for everyone. But you can't get there without protocols nobody can switch off, censor, or control. Self-custody is what makes that possible.

We still have a long way to go for it to be accessible for everyone. It should come in multiple forms for different use cases: from abstracting the technical complexity for low-balance transactional wallets, to the most secure multi-sig setups for long-term wealth. If you actually care about crypto's values, one of the highest-leverage things you can do is make self-custody better, safer, and more accessible.

At least that’s what I'm doing :)