Raoul Pal lays out a machine-economy thesis on X
Raoul Pal said on X that most economic activity could be handled by AI agents within the next two years, with humans no longer sitting inside the transaction flow in the way they do now. His argument is that autonomous agents will identify trades, borrow capital, put on positions, hedge risk, and complete settlement on their own, while the current banking system will not be able to support machine-to-machine activity taking place millions of times per second.
In the post, he wrote: “Billions of AI agents are about to start transacting millions of times a second. They can't use a bank… so the whole machine economy will settle on crypto rails instead.”
Pal described a transaction that, today, would usually involve one person initiating it, another reviewing it, someone recording it, and another party completing settlement. Strip the humans out, he said, and software agents can spot the opportunity, borrow funds to build a position, hedge with a second agent, and settle with a third. The sequence runs automatically and almost instantly. He said that kind of process will be repeated constantly by billions of agents, at a speed far beyond human perception.
Why he thinks the current human-centered economy is hitting a wall
Pal tied the shift to what he sees as a labor problem. Economic growth, he said, comes from only two sources: a larger workforce or better output from existing workers. Western economies had both for decades. In his telling, they now have neither.
He pointed to long-running declines in birth rates, a shrinking labor pool, and the weaker productive profile of aging societies compared with younger ones. Once growth fades, governments are left borrowing more and printing money to service debt, he argued. The result, in his view, is a steady decline in currency value, which shows up in cash losing purchasing power while housing and equities keep outrunning wages.
His conclusion is that governments cannot create workers who were never born. The alternative, he said, is to build a new labor force based on silicon rather than human labor.
Why banks do not fit machine-to-machine commerce
Pal’s argument turns on the idea that machine labor must be able to transact. Those transactions may need to happen millions of times per second, but the banking system was designed around humans and does not map cleanly onto software agents.
He first pointed to account opening. Banks verify identity and determine whether the applicant is a real natural person. Software does not meet those assumptions, so machines cannot simply plug into bank infrastructure as it exists now.
He then moved to payment granularity. According to Pal, banks generally operate with a minimum unit of 1 cent, while AI agents may need to pay for a single data query, a tiny slice of compute, or one call to another agent’s service. Those payments can fall well below 1 cent and occur millions of times per second. He said legacy finance cannot represent those amounts effectively, much less process that volume.
Speed is the third issue in his view. Cross-border wires often take days, pass through intermediary banks, incur fees at multiple points, and depend on manual records and reviews. The system also slows or stops outside business hours. That may be manageable for a human making only a few transfers each month. It does not work, Pal said, for machines that may need to complete thousands of transactions in the time it takes a person to read a sentence.
His case for blockchains as the settlement rail
Pal contrasted that with blockchain-based infrastructure. On suitable public chains, he said, a cross-border payment can settle in about 300 milliseconds from start to finish. No intermediary bank is required, no one has to wait for a business day, and a wallet can function as the identity layer.
He also stressed precision and programmability. Blockchain-based money can support decimal precision out to 18 places, which lets payments be split into very small amounts. The system runs all year without closing, and payment logic can be embedded directly into the asset itself. In the examples he gave, funds can be released only after work is delivered, routed automatically to 10 separate agents, or refunded if a condition is not met, all without human approval.
That is the core distinction in his framing. Blockchain is not just a faster bank. Banks move money slowly between humans on a limited schedule. Blockchains, as he described them, move value instantly between machines and can carry the transaction rules inside the payment itself.
Tokenization, in his view, is much broader than putting stocks on-chain
Pal said the real purpose of the crypto industry is not token prices but settlement infrastructure for machine economies. If machines need a place to clear value, he argued, blockchains are the viable route, and value will accumulate at the infrastructure layer that supports those systems.
He said many people still think of tokenization as tokenized stocks, tokenized bonds, or tokenized real estate. Those categories are real, but he called them only a narrow slice of the bigger picture. In his definition, tokenization is a packet of information that machines can read and use. That, he said, is one reason AI model tokens and blockchain tokens share the same term.
From there, he extended the concept to a much wider set of assets and rights. Dollars can become stablecoins. Identity can become a verifiable credential. Permissions can become keys. Information can become something directly bought and sold. Energy, storage, and compute can all be turned into assets that machines exchange instantly with one another.
He added that large sets of scientific files, climate records, soil samples, anonymized hospital cases, and farm sensor readings may not be worth much today. But if billions of intelligent agents need that data to make decisions, those datasets gain economic value. In that setup, agents pay for the data, the data becomes tokenized, and a new market appears where little or none existed before.
Pal grouped stablecoins, lending, real-world assets, storage, and identity under the same umbrella. In his telling, these are not separate narratives but different names for pieces of one machine system.
He says wages could lose their place as the main distribution mechanism
Pal pushed back on the idea that machine labor automatically means economic collapse. His argument is not that humans disappear from the economy, but that billions of new economic participants enter it. Those participants consume energy, compute, storage, data, and settlement capacity every second. In his view, demand does not collapse under that model. It rises sharply.
Where he does see a break is in the way humans get paid. Wages, he said, are essentially the price of human time, because labor has historically been the scarce resource in the economy. If human labor stops being scarce, wages stop functioning as a reliable way to distribute output. Retirement plans, mortgages, and long-term personal planning have all been built around that structure, he said, and now face competition from machine labor that runs on electricity.
Pal said he had discussed that human position before in an essay titled Economic Singularity. His short version here is that if intelligence becomes cheap and widely available, scarcity shifts toward specifically human qualities: trust between people, taste, and the authenticity that comes from face-to-face interaction. Machines may reproduce many outputs, but they cannot reproduce a human being whom another person is willing to trust.
Where he thinks value will flow
Even so, Pal said wealth will move toward machine owners and toward the infrastructure that keeps those machines running. He added that this moment differs from earlier periods because anyone can own part of that infrastructure. In his example, that is true whether someone lives in London or in a village where 10 people share one phone.
His closing point was that people do not need to beat machines on transaction speed or chase every invisible trade happening at machine scale. What matters, in his framing, is owning part of the underlying system and letting value accumulate there over time.