For the entire history of money, the thing spending it has been a person. A human decided, a human authorised, a human clicked. Every payment system ever built — cash, cheques, credit cards, SWIFT, the entire apparatus of modern finance — assumes a human being at the moment of decision. That assumption is now breaking. AI agents are becoming economic actors in their own right: buying compute and data and browser sessions, paying for services, and increasingly managing real money on behalf of the humans they serve. And they are discovering that the financial system built for people does not fit them at all.
This is the agentic economy, and it is arriving faster than its infrastructure. To understand it, you have to see three shifts at once: a change in what money is (tokenisation), a change in what money does (agentic delegation), and a change in what the world must provide for it to work (agent-ready infrastructure). Underneath all three sits a fourth question this blog has circled for many issues: what money do you anchor to when the money itself can act?
From Messaging to Movement: What Tokenisation Changes
Start with the deepest shift. The money systems that run the world today — SWIFT, Visa, Mastercard — are not money. They are messaging systems. When you pay across borders, SWIFT does not move your money; it sends a message telling banks what to do, and the actual settlement happens later, slowly, through chains of correspondent banks, separated from the asset itself. The instruction and the value are two different things travelling on two different tracks.
Tokenisation collapses that separation. With tokenised money — a stablecoin, a tokenised bank deposit, a central-bank digital currency — the token is the value. The asset and the instruction become the same object. And once they are the same object, things become possible the messaging world never could.
Atomic settlement means the money and the asset change hands in the same instant, eliminating the risk that one side pays and the other defaults. Programmability means conditions can be written directly into the money — pay only when the goods arrive, release funds only if a threshold is met, split a payment automatically among parties. This is the leap from static money to programmable money. And programmable money is the precondition for the next leap: money that acts on its own.
Agentic Money: Delegating the Decision
The financial writer Chris Skinner’s phrase for what comes next is agentic money. We are moving, he argues, from a world where humans authorise every transaction to one where we delegate that authority to systems. Money becomes proactive, not reactive. Within rules a human sets, machines transact continuously, in real time, with no person approving each step. This is less science-fiction than it sounds: algorithmic high-frequency trading already accounts for the majority of equity-market volume, executed entirely by machines with no human placing individual orders. Agentic money extends that principle from the trading floor to the whole of economic life.
We are moving from a world where humans authorise every transaction to one where we delegate authority to systems. Money starts to become agentic.
— after Chris Skinner
The Four Gaps Blockchains Close
Here the AI story and the crypto story, so often treated as separate hype cycles, turn out to be one story. The venture firm a16z argues that AI agents became economic actors faster than the infrastructure to support them could form. Agents are effectively unbanked — and blockchains provide exactly the primitives the agentic economy is missing. There are four gaps. Identity: agents have no portable, verifiable identity, so the proposed answer is “Know Your Agent” (KYA) — cryptographically signed credentials linking an agent to its principal, permissions, and reputation. Governance: when agents run real systems, the control problem becomes who controls the model weights? — an infrastructure problem requiring cryptographic proof of provenance, not a policy question. Payments: stablecoins are emerging as the settlement layer, and the dormant HTTP 402 (“Payment Required,” specified in 1997 and never used) is being revived as the x402 standard, letting an agent pay directly inside an HTTP request — crypto rails win because a headless merchant with no website and no legal entity is nearly impossible to underwrite traditionally but trivial to pay on permissionless rails. User control: scoped-delegation toolkits and agentic wallets with spending limits let a human draw cryptographic boundaries around what an agent may do.
For 300,000 years, human cognition was the bottleneck. As intelligence becomes cheap, verification becomes expensive — and the comparative advantage of being human moves up the stack.
— after a16z Crypto
That last line contains the deepest idea in the field. When a scarce resource — here, cognition — suddenly becomes abundant, the bottleneck moves. If intelligence is cheap, the new scarcity is verification: knowing that an agent did what it was supposed to, did not drift from your intent, did not optimise for the wrong proxy. Blockchains, with their auditable trails, make verification cheap again, and the human comparative advantage shifts upward — from doing the work to setting the direction and bearing the liability. That is the same conclusion the ikigai argument reached about meaning, arrived at here through economics.
The Web Has to Learn a New Language
None of this works if the web itself cannot talk to agents — and right now, it mostly cannot. When Cloudflare scanned the 200,000 most-visited domains for agent-readiness, the findings were stark: machine-usage signals on a handful of sites, clean machine-readable content on almost none, formal capability descriptions on fewer than fifteen sites in the entire world. The web learned to speak to browsers in the 1990s and to search engines in the 2000s. It is now being asked to speak to agents. The emerging standards are technical but the logic is simple: give machines a clean way to discover what a site offers (capability cards), read its content without human-oriented clutter (markdown negotiation cuts the tokens an agent must process by up to 80%), and pay for what they use (x402). The sites that adopt these standards now will be visible to the buyers of the next decade — buyers who are not people.
The Layer That Does Not Move
Step back and the full architecture comes into view — and it is a stack, not a single thing.
The upper layers are where the action is — agents transacting, stablecoins settling, smart-contract platforms providing the programmable rails. This is the flexible, programmable, agentic part of the system, and flexibility is exactly its nature: stablecoins can be issued and frozen, contracts can be upgraded. But a system made entirely of flexible, issuable, freezable money has no anchor. This is where Bitcoin’s role becomes clear — precisely because it is none of those things. It is fixed at 21 million, issued by no one, freezable by no one: the deliberately non-agentic layer, money whose scarcity is anchored in physics rather than promise. In an economy where money can increasingly act, the one money that simply is — that holds its meaning regardless of any agent, issuer, or instruction — becomes more valuable, not less. The agentic economy needs programmable money to act with, and sound money to anchor to. These are two different jobs, and conflating them is the central confusion of the field.
What It Means
Money is becoming infrastructure for machines, not just for people. Every payment system we have was designed around a human decision-maker, and that assumption is dissolving. When the buyer is an agent — transacting thousands of times a second, with no hands to click and no patience for three-day settlement — the human-centric rails simply do not work. The money that fits the agentic world is programmable, permissionless, and continuous. That is a description of crypto rails, which is why the AI and crypto stories are converging whether or not their loudest proponents realise it.
And the clearest way to think about the whole field is to separate two jobs. One job is to transact — flexibly, programmably, conditionally — and stablecoins and tokenised assets do that. The other job is to store value in something no agent, issuer, or state can alter — and only Bitcoin does that. The mistake everywhere in the discourse is to treat these as competitors. They are complements: an agentic economy that transacts at machine speed still needs a fixed point to measure against. The more the money on top can do, the more it matters that the money underneath can do nothing at all. Part 2 takes the same shift up a level — to geopolitics, and crypto as a neutral settlement layer between states.
Flight Log — Dispatch from Altitude
Modern aviation has already lived through the transition the economy is now facing: the move from a human authorising every action to a human supervising a system that acts on its own. My A320 flies most of the route through the flight management system — it navigates, holds altitude, manages thrust, and executes a programmed descent, all without my hand on the controls. I do not authorise each adjustment. I set the constraints, and the system acts continuously within them. This is, in the most literal sense, agentic flight.
And aviation learned, sometimes through tragedy, the exact lessons the agentic economy is about to learn. The first: delegation requires clearly drawn boundaries. The automation does precisely what it is told within its envelope, and the accidents happen at the edges — an ambiguous input, a mode the crew did not realise was active, an instruction that cascaded in a way no one intended. The agentic-money people call this “scoped delegation” and “ambiguous inputs triggering multi-step workflows.” We call it a mode-confusion accident, and we have written the rules in blood.
The second lesson: as the machine takes over execution, the human role moves up the stack — from flying to verifying. My job is no longer to be the fastest controller of the aircraft; the automation wins that contest. My job is to verify that the system is doing what I intended, to catch the drift between what I commanded and what it understood, and to hold final responsibility when it matters. That is exactly the shift the economists describe: when execution becomes cheap, verification becomes the scarce and valuable thing.
And the third lesson is the one about anchors. No matter how sophisticated the automation, there are fixed quantities it cannot argue with — the fuel in the tanks, the length of the runway, the laws of aerodynamics. The system can optimise everything else, but it cannot program away the physical constants. The agentic economy will be the same: layers of brilliant, programmable, self-acting money on top — and underneath, something fixed, that no agent can inflate or freeze, that the whole structure can be measured against. Bitcoin is the fuel gauge of the agentic economy: the one number that reports a physical reality the system cannot talk its way around.
I have flown the agentic future for my entire career. It works — brilliantly — but only because someone drew the boundaries, kept a human verifying at the top, and never let the automation forget the fixed quantities at the bottom. The economy is now boarding the same aircraft. The question is whether it has read the accident reports first.