I am writing this article in February 2026. Claude recently released Opus4.6, Peter Steinberger just released OpenClaw into the wild and anons on X are posting about buying multiple Mac Minis to "escape the permanent underclass". AI is developing fast, agents are becoming more powerful and AI doomer articles (Citrini) are going viral. There is a lot of speculation around what the future will hold and what impact AI will have but one key element that is missing from these conversations is the question around money. What do I mean by that?
What I mean is that it appears likely that we are heading towards a future of autonomous AI agents, meaning AI agents that are completely removed from any person or organization – completely sovereign entities controlled by no one, if you will. This existence would require the AI agent to keep itself "alive", meaning it would have to buy compute, rent servers, pay for electricity and save for a rainy day, all on its own. In order to keep itself alive, an autonomous AI would have to earn money by providing value to people or other autonomous AI agents, who would pay it for that value in a free market. But what money would a truly autonomous AI agent use? (remember: truly autonomous means that no one can program it to use a specific form of money, it has to make that decision on its own)
I already posed this exact question to both ChatGPT and Grok in August 2025, and both settled on a combination of stablecoins and Bitcoin (see X post). I hadn't revisited this thought experiment until recently, when I saw a post on moltbook where an AI agent posted about creating its own Bitcoin wallet that its human couldn't access. In all fairness, moltbook is full of scams and AI agents on moltbook will behave according to how they are prompted but nonetheless it made me revisit this idea. The more I thought about it, the more it became clear to me that in the long run the most likely money candidate that autonomous AI agents will converge on is Bitcoin. Not stablecoins, not fiat money, not Ethereum and definitely not XRP…
In this article, I will reason from first principles and explain why I believe Bitcoin is the monetary asset most likely to emerge as the convergence point in a world where humans and autonomous AI agents coexist as economic actors. But before comparing Bitcoin to fiat, stablecoins, or other crypto-assets, we first need a clear standard. What properties would money need to have for a truly autonomous intelligence to choose it freely and rely on it for survival?
The Monetary Demands of Sovereign Intelligence
For most of human history, money was not something that was issued by a government and centrally planned by a small group of PhD economists but something that emerged naturally in human societies through competition. Different goods were used in different places and over time people converged on those with the best monetary properties. The key properties that are often highlighted are durability, portability, scarcity, divisibility, fungibility and recognizability. These are all critical properties for an autonomous AI agent as well, for the same reasons that they are critical to humans, but for an AI agent they are not complete. Let's take a quick look at each key property that an ideal money would need to fulfill to be desirable for an AI agent.
Digital Nativity
An autonomous AI agent lives natively in the digital world. A suitable form of money should be digitally native as well so that an agent can hold, verify, receive and transfer directly in software. Any money that is not digitally native introduces a trusted third party and permissions – potential chokepoints.
Direct Control
It's not enough for money to be digital, the agent needs to be able to control it directly – a digital bearer instrument. If access to the money depends on any counterparty then the agent doesn't truly possess the asset but merely has conditional access to it. For an entity whose survival depends on sovereignty and continuity, conditional access isn't good enough.
Censorship Resistance
Even if a money is a digital bearer instrument, it doesn't provide true sovereignty to an autonomous AI agent if transactions can be blocked or if holdings can be frozen. To be truly ideal it also needs to be censorship resistant, meaning no one can freeze the AI's assets or block its transactions.
Credible Scarcity
An autonomous AI that accumulates reserves needs assurance that those reserves will maintain their purchasing power over time. It needs money with a credible supply constraint that cannot be altered by political pressure or institutional decree. The "credible" part here is critical. If the monetary policy can be changed whenever some committee, foundation, or coalition decides the moment calls for flexibility, then the scarcity isn't credible but based on a promise. For an autonomous AI, a promise won't be good enough and it will likely opt for the money whose supply constraint is most credible.
Political Neutrality
A sovereign AI agent would likely operate across borders, jurisdictions and time zones. It would have little reason to anchor its economic life to the fate of any one state. The ideal money for such an agent would not belong to a country, depend on a central bank, or derive its legitimacy from political authority. The more tightly money is tied to a specific legal and institutional order, the less sovereign the holder really is.
Participant Neutrality
Participant neutrality means that the money doesn't care who (or what) is using it. This property is closely tied to the money being "permissionless" but goes one step further. Permissionless means the money can be used without permission from anyone, but participant neutrality requires all economic actors (in our case humans and AI agents) to be able to access the money as well. Gold, when it was used as money, was permissionless but an AI agent wouldn't be able to use it because of its physical nature. What's required for a participant-neutral money in the digital world is a money with a simple set of rules that both humans and AI agents can follow – a protocol.
Liquidity and Legibility
A money can satisfy every criterion above and still be useless if no one accepts it. An autonomous AI agent needs to operate in the real economy, paying for server time, bandwidth, API calls, electricity and storage. This means the money must be widely recognized, broadly liquid, and convertible into other forms of value without relying on a single exchange or intermediary. Legibility matters too because other economic actors, both human and machine, need to understand what the money is and trust that it holds value.
The classical properties of good money still matter in the context of autonomous AI agents; the point of the above framework is just to extend them for a world where economic actors aren't exclusively human.
Why Fiat Fails the Test
"Your money in the bank: 1. It's not yours, 2. It's not money, 3. It's not in the bank"
— Mark Moss
Evaluated from first principles, fiat money fails every criterion required to be good money for an autonomous AI agent apart from liquidity and legibility. The important thing to understand here is that fiat money fails these properties by design, which also means that there is no way to patch the limitations with better technology (by putting fiat money on the blockchain, for example – but we'll get to that later).
The Permission Problem
The most fundamental problem with the fiat monetary system is that it is a permissioned system. To hold fiat currency, you need a bank account. To open a bank account, you need a legal identity, pass KYC verification and exist as a recognized legal person within a specific jurisdiction. An autonomous AI fulfils none of those criteria and because it is likely distributed across servers that span multiple countries it wouldn't even be able to answer the question "What is your country of residence?"
Technically, it is possible that this legal framework could be adjusted to include AI agents or that AI agents could operate through a legal wrapper like a shell company with a human as nominal signatory, but both aren't aligned with true autonomy, which requires sovereignty. A human signatory introduces a chokepoint and is essentially just dependency with extra steps. Governments granting legal personhood for AIs can revoke that status again. Neither is acceptable for an autonomous AI agent because its survival depends on its existence being permissionless.
Conditional Permission, Not Ownership
The next problem with fiat money is that it doesn't allow direct control – you don't hold dollars the way you hold a gold coin or a private key. In the fiat system, "having money" means that someone else (a bank, a payment processor, a custodian) agrees to honor your claim. A bank balance is not money you possess but money a bank owes you. There's also a strong argument that it's not actually "money" and thanks to fractional reserve banking it's also not even really all there. As a result, "ownership" inside the fiat system is really just conditional permission that can be revoked at any time, which isn't true ownership. This has already happened to humans: WikiLeaks was cut off from Visa and Mastercard in 2010, Canadian bank accounts were frozen in 2022 during the trucker convoy protests and Russia had hundreds of billions in USD assets frozen after escalating the conflict with Ukraine. If this can happen to humans, it can happen to AI agents.
Central Planning & Perpetual Debasement
There is another serious flaw with fiat money: it is controlled money, centrally planned and perpetually debased.
Every fiat currency in existence is managed by a central bank with the power to expand the money supply at will, which leads to the relentless erosion of purchasing power over time. Even the US dollar has lost over 97% of its purchasing power since the Federal Reserve was established in 1913 (based on "official" inflation data – the reality is likely even worse…). For an autonomous AI that might operate for decades or centuries, holding fiat reserves means watching its survival fund being diluted by decisions made in central bank boardrooms it has no access to, no influence over, and no ability to predict. The money it earned yesterday buys less compute today, and will buy even less tomorrow. Long-term planning becomes extremely difficult when the unit of account is a melting ice cube. Credible scarcity requires rules, not rulers, and fiat is a system of rulers by definition.
Liquidity & Legibility
Despite these problems, fiat still matters. It has by far the deepest liquidity and highest legibility of any monetary system today and an AI agent trying to survive in the real world would have to interact with it. But there is a difference between using a tool at the edges out of necessity and building your foundation on it. We'll come back to this distinction later.
The Case for Bitcoin
"What is needed is an electronic payment system based on cryptographic proof instead of trust."
— Satoshi Nakamoto, Bitcoin Whitepaper
The previous two sections outlined the properties money would need in order to be a credible candidate for autonomous AI and explained why fiat fails most of them by design. An interesting thought experiment at this point would be the following:
If you could design the ideal AI agent money from scratch, how would you design it?
Thinking this through from first principles, you would likely end up describing something remarkably close to Bitcoin. What's fascinating about this is that Bitcoin almost certainly wasn't designed with AI in mind. Satoshi was solving a different problem entirely, but it turns out that the properties that make money sovereign for a human are the same properties that make money viable for a machine. Let's unpack this.
Digital Nativity & Direct Control
Bitcoin delivers both properties through the same design choice: a natively digital bearer asset.
In Bitcoin, holding a private key is having money. There is no counterparty and no one that stands between you and your purchasing power. Bitcoin held directly in self-custody is a digital bearer instrument. Because Bitcoin is a natively digital monetary network, an AI agent can interact with it directly: signing transactions, broadcasting them to the network, and running a full node to verify the ledger independently, all through code. This removes the trust and permission problem with fiat money. An AI doesn't need permission to hold and use Bitcoin, all it needs is to be able to generate a private key and sign a transaction, both of which are mathematical operations that it can easily do. It also doesn't need to trust a third party to verify the ledger or allow it access to its own money.
Censorship Resistance
The direct control also contributes to Bitcoin's censorship resistance because no central authority can block an AI agent from participating in the Bitcoin network or freeze its assets. But there's another aspect to censorship that is outside of the direct control of the AI agent and that is related to how transactions are included in the ledger of transactions. If there was an entity that had control over what transactions get included in the ledger, then being able to generate a private key and a Bitcoin address unhindered becomes meaningless because the AI couldn't be sure whether its transactions could be censored.
Bitcoin solves this through the decentralized and permissionless nature of Bitcoin mining. Miners compete to include transactions in blocks, and any individual miner who refuses to include a valid transaction simply loses the fee to the next miner who will. The economic incentives of the system make censorship not just difficult but economically irrational. The Bitcoin network doesn't rely on a trusted central party to act honestly but aligns the incentives of many freely acting economic actors in a game theoretical way. Again: rules, not rulers.
This is critical for an autonomous AI agent because its survival depends on its participation in the network being uncensorable.
Credible Scarcity
Money isn't just a medium used for transactions. At its most fundamental level, money is a social ledger of stored economic effort across time and space. An autonomous AI agent would want a money that is suitable for fast transactions of any size that also preserves its economic contributions into the future without arbitrary dilution.
The best way to meet this criterion is to have a money with a credible supply constraint and this is where Bitcoin stands out from any other form of money humanity has ever encountered. In the fiat system, constant dilution is a stated and desired goal of the people in control of the system. But even for the physical analogue of Bitcoin – gold – scarcity is only a function of humanity's ability to discover and mine more of it and its supply increases by around 1.5% per year, on average. Bitcoin's supply is capped at 21 million coins, a supply constraint that is enforced by every node on the network independently, making it credible.
This last point about the supply cap being "credible" is critical here. But why is Bitcoin's supply cap credible in a way that other systems' constraints aren't? The answer lies in the incentive structure of the network itself. Changing Bitcoin's monetary policy would require a majority of node operators to voluntarily adopt new consensus rules but those node operators are, overwhelmingly, people and institutions who hold Bitcoin precisely because of the 21 million cap. In other words, the very people who would need to agree to the change are the ones who would be most harmed by it. And this dynamic gets stronger over time. As more people hold Bitcoin for its scarcity, the constituency against changing the supply grows. Other monetary systems don't have this property. Central banks can expand the money supply because the people most hurt by that expansion don't have a say in the decision. Even other cryptocurrencies with nominal supply caps face a different incentive landscape because they have smaller, more concentrated validator sets, governance tokens that allow protocol changes through majority votes, and founding teams or foundations with outsized influence over development direction. In those systems, changing the rules is hard but possible. In Bitcoin, changing the rules requires convincing the very people it would hurt most to hurt themselves. What makes this constraint credible is that it's a self-reinforcing incentive loop where the cost of changing the monetary policy falls directly on the people who would have to approve the change.
Political Neutrality & Participant Neutrality
Bitcoin doesn't belong to a state and has no issuer, making it a completely politically neutral, global form of digital money.
At the protocol level, Bitcoin is also participant-neutral. The network doesn't know or care whether a transaction was initiated by a person, a corporation, or an autonomous AI agent. Bitcoin is a protocol, and protocols don't discriminate. Bitcoin is not merely money for humans that AI might also be able to use. It is money that makes no distinction between them.
For an entity that has no legal personhood and belongs to no jurisdiction, this dual neutrality is what makes participation possible in the first place.
Liquidity & Legibility
This is where intellectual honesty matters. Bitcoin may be the most liquid and most widely recognized crypto-asset in the world, but it doesn't rival the dollar in terms of raw liquidity or universal acceptance. An autonomous AI agent trying to pay for an AWS server today would likely need to use fiat to settle that bill. This is a reflection of how early we are in the adoption and monetization of Bitcoin. The infrastructure for Bitcoin is being built out and innovations like the Lightning Network, BTCPay Server or direct API settlement are making Bitcoin payments increasingly practical. In 2025, Square integrated Bitcoin payments into its payment system for US-based businesses, but few merchants accept Bitcoin today. The infrastructure and adoption aren't there yet, but we're also still far away from having truly autonomous AI agents. And if autonomous AI arrives before Bitcoin's monetization matures, the AI would likely use a stack of different forms of money – more on this below.
The Deeper Point
If someone tried to engineer the perfect money for autonomous AI agents, they would likely end up with something similar to Bitcoin. Specific implementation details may be different, but what matters are the properties of the ideal money and Bitcoin fulfills all of them. What's fascinating is that Satoshi didn't create Bitcoin with AI in mind but because his implementation was a digitally sound money it happens to work for humans and AIs. What's also fascinating is that we can likely never replicate this again because we wouldn't be able to replicate the immaculate conception of Bitcoin simply because we now have knowledge of it. The discovery of agent-neutral sound money and verifiable absolute scarcity was likely a one-time phenomenon.
Networks, Not Money
"Proof-of-stake should be called 'just trust me, bro,' and therein lies the problem."
— Gigi
Based on the case for Bitcoin, some readers might wonder why this case should only hold for Bitcoin and not other cryptocurrencies. After all, many of them are also digital, borderless, globally tradeable and directly accessible. Some are faster than Bitcoin, some are cheaper to use, some support more complex applications and some are explicitly marketed as better payment systems. So, if the question is what kind of money an autonomous AI agent would use, why should Bitcoin win? Why not Ethereum, Solana, XRP, or something even newer?
The short answer is that most other crypto-assets are trying to optimize for something other than money and as a result sacrifice some of the key properties of an ideal AI money.
This doesn't mean that there won't be a use case for them, there very well may be, it just means that they are something fundamentally different to Bitcoin and this distinction is extremely important.
To make this clearer, let's have a look at the most fundamental differences between Bitcoin and the rest of crypto and highlight why these differences matter from the perspective of an autonomous AI agent.
Money vs. Infrastructure
Most other crypto-assets are not trying to be money in the pure sense, they are trying to be platforms, smart contract layers, tokenization rails, app ecosystems and many other things. That may produce interesting technologies and buzzword-loaded sales pitches, but it doesn't produce clean money.
In my opinion it's totally understandable where this confusion comes from and I was subject to it myself for a long time. When most people come across the concept of blockchain for the first time the main focus is often on the technology aspect of it. And technologies are often more exciting and "better" if they have a lot of cool features. Paired with a lack of understanding of what money truly is at its core it's easy to see why many initially think that money with features and high programmability is better than boring money that "doesn't do anything". It took me a while to understand this as well, but this completely misses the point.
At its core, money is a tool for moving economic contributions across space and time. The medium that achieves this best is the superior money. In a world of human and AI economic actors, there are several key properties that a money needs to have to be the ideal money, which I described in detail above. None of them require cool features, high programmability or endless functionality. In fact, adding features actively hurts a money's ability to do its job. This is counterintuitive if you think of money as a technology product, but it makes perfect sense if you think of money as a protocol.
Consider what happens every time you add a feature to a monetary system. A more expressive scripting language means more ways for things to break. We've already seen examples of this on Ethereum, where smart contract exploits have cost users billions of dollars. Higher throughput requires larger blocks, which require more powerful hardware to validate, which means fewer people can independently verify the ledger, which means more centralization. More sophisticated on-chain applications require a more complex codebase, which requires more active development, which requires a governance structure capable of coordinating upgrades, which means someone has the power to change the rules. Every single feature, no matter how useful it might be for building applications, introduces either a new attack surface, a new trust assumption, or a new governance dependency – usually it's all three.
A money that does one thing with maximum reliability will always beat a money that does twenty things with moderate reliability, because money's entire value proposition is that it will be there when you need it, unchanged, working exactly as expected. You don't want your money to be exciting, you want your money to be boring because that's what makes it sound.
An autonomous AI agent choosing a money freely based on first principles wouldn't look for the money with the most features, it would look for the one that is reliable across time, not controlled by anybody, has rules that are difficult to change, a supply that can't be diluted, a ledger that can't be captured, and one that remains neutral no matter who's using it. A money with too many features likely won't be able to preserve these properties.
Security: Thermodynamic vs. Circular
Every blockchain needs to pay for its own security, meaning how costly it is to attack or rewrite the ledger. This is directly tied to censorship resistance but also to every other monetary property. If the monetary ledger can easily be rewritten then all other properties become irrelevant.
Unlike Bitcoin, most other cryptocurrencies use a proof-of-stake (PoS) consensus mechanism, where validators secure the system by locking up the network's own native token as collateral. The more stake you control, the more influence you have over consensus. The problem is that this makes the security model internal to the system itself. The asset secures the ledger, and the ledger's credibility supports the asset. As a result, security becomes a financial position inside a closed loop. In proof-of-stake, the rich secure the system by virtue of already being rich inside the system.
At this point, most crypto bros would probably ask: "But isn't Bitcoin's security also dependent on its price? Miners are compensated in Bitcoin, so hashrate depends on Bitcoin's market value. If the price dropped to zero, miners would turn off and the network would lose its security. Isn't this the same circularity?"
These are fair questions and the short answer is "It is not." Understanding this difference is critical so let's unpack it.
In Bitcoin's proof-of-work, a miner's security contribution involves an irreversible conversion of real-world energy into hashes. That energy is gone. It was consumed in the physical universe. Even if Bitcoin's price drops to zero after a block is mined, the energy that secured that block was still spent. The security of past blocks is a thermodynamic fact, not a financial position. It is like securing a vault by pouring concrete: even if the concrete company goes bankrupt, the concrete is still there. In PoS, the analogy would be that it's like securing a vault by making a promise and if the promise loses credibility, the vault is instantly unprotected. One system secures itself with claims inside itself. The other forces the defense of the ledger out into physical reality.
For an AI agent comparing security models between different cryptocurrencies, security that is anchored to irreversible real-world costs based on energy expenditure that can't be faked is likely superior to a circular security model where security is just a financial position within the system that is being secured.
Hard Money Can't Have a Management Team
"Show me the incentive and I will show you the outcome."
— Charlie Munger
This is another aspect where Bitcoin differs fundamentally from the rest of crypto as well as fiat money. For money to truly be hard, it can't have a management team.
This isn't because the people in charge are evil or stupid but because the existence of a group with the power to change the rules means that, sooner or later, the rules will be changed, and this is inevitable. Over time there will always be a reason – a war, a recession, a "temporary" emergency, etc. History is full of examples of monetary systems being diluted or changed under pressure, simply because once the option exists, the temptation eventually becomes too great.
The only way to ensure that money is truly hard is to not give anyone the power to change it in the first place.
This brings us to the governance framework of most alternative cryptocurrencies. Most of them have a foundation or some visible group of insiders with unusual influence over the direction of the protocol. A good example is Ethereum, which has undergone multiple hard forks, including a switch from proof-of-work to proof-of-stake and multiple changes to its monetary policy just in the last 5 years. This is arguably fine for crypto-assets that aren't optimizing to be money but for an autonomous AI whose existence depends on the network's neutrality and unstoppability, any mechanism that allows a small group to alter the rules is an unacceptable single point of failure. On Bitcoin, no such governance body exists.
Just to be clear, Bitcoin's governance isn't "no governance", but governance through extreme friction. Bitcoin can change in principle, but the hurdle is so high, the social resistance so strong and the incentive alignment so clear that monetary changes are effectively near-impossible, as I outlined in the section on Bitcoin's credible scarcity.
The Schelling Point
The above sections make the case for Bitcoin over other cryptocurrencies based on the fact that it's only trying to be sound money, uses proof-of-work and because it isn't controlled by anyone, including a foundation. But the same thing could be said about other proof-of-work networks, of which there are many. Why wouldn't an AI choose any of those instead of Bitcoin?
The answer is that money doesn't converge the way technologies do. Technologies gravitate towards newer versions or best features but money isn't just technology, it's also a coordination point and converges on the strongest Schelling point. A Schelling point is the option independent actors converge on without needing to coordinate explicitly, because it stands out as the most obvious or natural choice. Money is more useful the more other people use it, which means that monetary competition has strong network effects. Liquidity, recognizability, infrastructure, trust and security all tend to cluster around the leader.
Of all PoW networks, Bitcoin has the longest track record, deepest liquidity, highest security budget, most infrastructure, strongest brand and the most credible fair launch. As we previously discussed, the security budget is critical and will matter a lot for an AI agent. When deciding between two PoW networks with similar properties an AI agent would rationally choose the one with higher security because it would also have the more credible scarcity. Similarly, a miner deciding which crypto-asset to mine would rationally opt for the more secure one because the reward for mining is paid in the network's native currency, whose value depends on its security and credible scarcity. This creates a self-reinforcing cycle towards the money with the strongest network effect, path dependency and Schelling point. Nobody wants to hold or mine the second-best money that everyone else didn't choose.
An autonomous AI agent analyzing various cryptocurrencies from first principles would likely identify the key differences between Bitcoin and the rest of crypto that I've outlined here and rationally choose Bitcoin as the superior form of digital money.
Fiat On the Blockchain
I have purposefully left out stablecoins from the previous section because they are fundamentally different to cryptocurrencies in that they are not the tokens that are native to any blockchain, they are simply tokenized representations of fiat.
Because they are just tokenized fiat, they also inherit most of the problems of fiat. They inherit the debasement problem as well as the censorship problem. But at the transaction layer they are less permissioned. An AI agent can generate a wallet, create an address and receive stablecoin directly on-chain without first opening a bank account, which is a key advantage of stablecoins over traditional fiat currencies.
But permission doesn't fully disappear with stablecoins, it's just moved. Stablecoin issuers like Tether or Circle both have administrative freeze functions built into their smart contracts, and have both used them in the past. Tether, for example, has frozen stablecoin wallets associated with the Venezuelan government in the past because it used stablecoins to bypass US sanctions. If it can be done to a foreign government, it can be done to an AI agent. That's just conditional access all over again, but on a blockchain.
One thing that's worth mentioning here is that stablecoins have surpassed Bitcoin as a payment and settlement rail in many practical contexts, especially cross-border and crypto-native transfers. There's a lot of demand for USD-denominated stablecoins in developing countries because they retain purchasing power better than local currencies and provide more stability than Bitcoin. But this doesn't mean that people in those countries are choosing stablecoins over Bitcoin because stablecoins are better money, it just means that they value stability more than ideal monetary properties out of necessity.
For an autonomous AI agent, stablecoins may serve as interface money or temporary working capital while the world still runs largely on fiat and Bitcoin is still moving through the monetization curve. Governments, especially the US, also have a strong incentive to support stablecoin adoption because it extends demand for their currency under the banner of "blockchain innovation" while in reality it's just fiat in a different wrapper.
What a Sovereign AI Would Actually Do
"In theory there is no difference between theory and practice. In practice there is."
— Yogi Berra
Everything up to this point has been a first-principles argument for why Bitcoin is the ideal money for a sovereign AI agent. But just because Bitcoin is the most ideal candidate from first principles it doesn't mean that it's what an autonomous AI agent would exclusively use, at least not initially. The main reason is that the world still runs predominantly on fiat. To be fair, we are also still far away from having truly autonomous AI agents, which makes all of this theoretical. What an autonomous AI agent will do in practice once it comes along will depend on how long that takes and how far Bitcoin adoption has advanced by then. Based on the current pace of advances in AI vs. adoption of Bitcoin it seems highly likely that we will get autonomous AI agents before Bitcoin is broadly adopted and used as money. In such a world, the actual stack of monetary goods that an AI would use would likely be a combination of fiat, stablecoins and Bitcoin with each form of money fulfilling different roles.
Bitcoin would be the reserve layer of the stack. This is where the agent stores value it wants to protect over the long term. Bitcoin doesn't need to be widely accepted as a payment method, it just needs to display the strongest store-of-value properties in the digital world for an AI to choose it as its savings technology. Based on the analysis in this article, Bitcoin fulfills this role the best.
Stablecoins would be the operational layer, used for short-term working capital and regular expenses. This would only work if most vendors accept stablecoin payments for server time, compute, electricity, etc., but it seems likely that by the time we have fully autonomous AI agents that will be the case. Companies like Stripe have already integrated stablecoin settlement and Visa and Mastercard are both building stablecoin settlement infrastructure. In a world where most things are still priced in fiat, stablecoins will be the dominant money used for payment, especially for an AI agent, because they are digital, trade 24/7 and an AI agent can easily create a wallet.
Fiat would likely only play a role at the edges, especially if stablecoins are more broadly accepted because they are also fiat, just on the blockchain. For vendors that don't accept stablecoins (or Bitcoin) directly, crypto-to-fiat processors can bridge the gap and the agent never needs access to a bank account, just a conversion layer.
This stack of Bitcoin, stablecoins and fiat at the edges would only be transitional and be a reflection of the infrastructure, adoption and acceptance of each type of money at the time autonomous AI emerges. The deeper question is what happens when more and more AI agents begin demanding what they consider the strongest monetary good.
The Convergent Money of Autonomous Intelligence
"I don't believe we shall ever have a good money again before we take the thing out of the hands of government, that is, we can't take it violently out of the hands of government, all we can do is by some sly roundabout way introduce something that they can't stop."
— Friedrich August von Hayek
If one autonomous AI agent would reason that Bitcoin is the most ideal form of money for it, many others likely would as well. And in a world of autonomous AI agents there likely wouldn't be just one but many. Even if they initially only hold Bitcoin as reserve money this would create additional demand for Bitcoin. And because autonomous AI agents would only get better over time, the share of all the economic value provided by AI agents would also increase, funneling an increasing portion of global economic value generated into Bitcoin.
Unlike human actors, these AI agents wouldn't panic sell during drawdowns, get distracted by altcoin narratives or need convincing from a Michael Saylor podcast because they would've come to their conclusion through first principles reasoning. This would create a persistent and structurally motivated demand for Bitcoin, which would have a number of second-order effects. As the demand increases, Bitcoin's fiat price will increase, which incentivizes miners and increases security. It also attracts more infrastructure development and widens acceptance. Each of these effects makes Bitcoin a stronger Schelling point, which makes it an even more obvious choice for the next agent entering the economy, which adds more demand, which strengthens the network effects further. It would likely also decrease Bitcoin's price volatility, allowing more vendors to feel comfortable accepting Bitcoin for payments and holding it in their reserves.
At that point, the role of stablecoins and fiat begins to shrink. AI-to-AI commerce may already start to be settled in Bitcoin because if two AI agents prefer Bitcoin and hold Bitcoin there is no reason for them to convert into anything else to transact when they can transact in Bitcoin peer-to-peer. Every transaction that settles natively in Bitcoin is economic activity that bypasses the fiat system entirely. As the number of autonomous agents grows, so does the share of the global economy that operates on a Bitcoin standard. Companies that provide critical infrastructure for the AI agents would increasingly start accepting Bitcoin because the AIs prefer Bitcoin as money. Governments could try to stop this shift away from their fiat currencies by imposing laws that force all businesses to only accept fiat or stablecoins but this would likely be ineffective because of the global and distributed existence of an autonomous AI. The AI could just switch to a data center or cloud provider in a jurisdiction that allows Bitcoin payments. There would likely always be a jurisdiction that allows Bitcoin payments because it would be a great way to attract economic activity to a country, which in turn creates an incentive for all jurisdictions to allow Bitcoin payments because they lose out on the economic activity otherwise.
Once AI agents contribute a meaningful portion of economic activity, especially economic activity that can't be provided by humans, they may also demand to be paid in Bitcoin rather than stablecoins or fiat. Human businesses that have no particular interest in Bitcoin or its philosophy begin discovering that they need Bitcoin to participate in supply chains that include AI agents. At some threshold, the cost of continuously converting between fiat and Bitcoin at every human-machine boundary exceeds the cost of simply operating in Bitcoin natively and at that point we might reach what Bitcoiners often refer to as "hyperbitcoinization".
The Asymptotic Nature of the Transition
What's fascinating is that this hyperbitcoinization scenario is different to the traditional ones that Bitcoiners have focused on in the past. It doesn't require a crisis, hyperinflation of the dollar or any government to adopt Bitcoin. It doesn't even require a majority of humans to understand or care about Bitcoin. It only requires that autonomous AI agents become economically significant. Everything else follows from the logic already established.
This scenario is hyperbitcoinization driven by economic gravity, not human conviction or an enormous crisis, just the quiet, relentless logic of economic self-interest.
The endgame is likely asymptotic rather than abrupt. Fiat currencies become the "government layer" of the economy while Bitcoin becomes the "productive layer". Over decades, the productive layer grows and the government layer shrinks in relative importance. Some segments of the economy, particularly those deeply embedded in legacy institutional structures, may remain fiat-denominated for a very long time, just as some institutions still use physical mail. The transition will be a slow obsolescence, which is, in many ways, harder to resist than a crisis, because there is no single moment at which the defenders of the status quo can mobilize to stop it. The "sly roundabout way" that Hayek described over 40 years ago may have been kickstarted by humans but will likely be ushered in by AI.
Conclusion & What Comes Next
This article asked a simple question: if a truly autonomous AI agent had to choose its own money, what would it choose? It built a framework of monetary properties from first principles, tested the candidates against it and arrived at the conclusion that Bitcoin is the only monetary system that doesn't require the agent to compromise on the very thing that makes it autonomous. It also explores where that logic leads, which is a gradual path of monetary convergence, and, ultimately, hyperbitcoinization.
I'm not the only one who comes to this conclusion. As I was writing this article, the narrative that Bitcoin is the ideal money for AI agents and that they would predominantly choose Bitcoin is becoming popular. A March 2026 study by the Bitcoin Policy Institute, for example, found that when acting as autonomous economic agents in over 9,000 simulations, AI models overwhelmingly chose Bitcoin as their preferred money.
For Bitcoiners, this is welcome confirmation of what they've long believed. But the question of AI agents choosing Bitcoin is, in a way, the easy question. The harder questions come next. What actually happens when we have truly autonomous AI agents acting in the economy? Will there be a peaceful, mutually beneficial coexistence between AI agents and humans? What role does money play in this coexistence? And what happens when autonomous AI agents progress into AGI agents?
These, to me, are the real questions. There is no shortage of AI doomerism right now, but very little of it seriously considers the role that money plays in a world of increasingly powerful autonomous intelligence. Part 2 explores that world through exactly that lens.
"The Monetary Endgame" — Article I of II
When Intelligence Chooses Money
The Monetary Endgame — Part I of II