A medium of exchange is a choice, not a law of nature, and it is a choice with a specific, nameable cost: every currency and every ledger you introduce is a surface someone can attack.
Somewhere in the United States today, a woman will donate a kidney to a stranger she has never met, because that stranger's brother is, at the same hour, donating a kidney to her husband. Her husband needed a kidney; she was willing to give him one and turned out to be incompatible. The stranger's brother was in the mirror-image bind. So a matching program paired the two families and swapped the donors. Four people go into operating rooms, two kidneys change hands, two lives are saved, and here is the part that should stop you: no money changes hands, no price is quoted, no credit is recorded, and no one owes anyone anything when it is over. It is a completed market transaction with no medium of exchange and no ledger of any kind.
That is worth dwelling on, because "build a marketplace with no money, no tokens, no credits, no barter accounting, no ledger" reads like a puzzle you'd set to be clever, a constraint so total it must be impossible or useless. It is neither. It is the kidney exchange, and it is not a thought experiment. It is operational national infrastructure, running in real hospitals, and in 2024 the United States performed more than forty-eight thousand organ transplants, a growing share of the living-donor kidneys among them arranged through exactly this mechanism. The economics of how it works won a Nobel Prize, and it turns out to be the cleanest possible answer to a question agent-marketplace designers are about to spend the next decade getting wrong.
Start with why money is there in the first place, because you cannot remove a thing well until you know its job. Money has exactly one job, and the economist William Stanley Jevons named it in 1875, in a book called Money and the Mechanism of Exchange. He called the problem the double coincidence of wants. In pure barter, for a trade to happen, I must have what you want and you must have what I want, at the same moment. That coincidence is rare. If I have wheat and want shoes, I have to find not just anyone with shoes, but someone with shoes who happens to want wheat. Money dissolves the coincidence: I sell wheat to whoever wants wheat, take the money, and buy shoes from whoever has shoes. The medium of exchange is a universal solvent for the matching problem.
Which means the instant you ban money, the double coincidence of wants comes straight back. This is the trap that sinks most "moneyless marketplace" designs before they start: they quietly become barter, and barter is worse than the thing you removed, because now every trade needs that rare simultaneous match. A moneyless market cannot be barter with extra steps. It has to solve the coincidence problem some other way.
And there is another way, one that most people never consider because money made it unnecessary for three thousand years. You do not need to find a pair who match. You need to find a cycle. I want your item, you want her item, she wants mine. No two of us can trade bilaterally, none of us has what the person we want to trade with wants, and yet the three of us together form a loop that clears perfectly. Everyone points at the thing they want, the arrows come back around, and all three swaps execute at once. No money moved. No debt was created. The cycle is the clearing. A marketplace with no money and no ledger is, precisely and formally, a problem of finding cycles in a graph of who wants what.
This is not hand-waving. It is a named, proven algorithm, and it is the intellectual core of the 2012 Nobel Prize in Economics, awarded to Alvin Roth and Lloyd Shapley "for the theory of stable allocations and the practice of market design." The specific mechanism comes from a 1974 paper by Shapley and Herbert Scarf on what they called the housing market, a toy economy where each person owns one house, wants exactly one house, and has strict preferences over all of them. The algorithm that solves it is called Top Trading Cycles, and Shapley and Scarf credited it to the mathematician David Gale.
It works like this. Every agent points to the holder of its single most-wanted item. Follow the arrows. In any such graph, where every node has exactly one outgoing arrow, there must be at least one cycle, a loop that closes on itself. This is a hard mathematical fact, not a hope: a finite graph where everyone points at someone cannot avoid a loop forever. You execute every cycle you find simultaneously, everyone in a loop gets their top choice and hands over what they held, you remove those satisfied agents from the graph, and you repeat with everyone left. The pointers redraw, new cycles appear, you clear them, and you continue until no one remains.
Look at what that mechanism does and does not need. It needs no money; nothing is priced. It needs no ledger; nothing is stored between rounds, because each cleared cycle is final and complete at the instant it executes. Nobody walks away holding a credit, a token, or an IOU. And the properties are not merely convenient, they are optimal in a way you rarely get in economics. On the strict-preference housing market it was built for, Top Trading Cycles is the unique mechanism that is at once individually rational, Pareto-efficient, and strategy-proof, and its outcome is the unique allocation in the core. Strategy-proof is the word that should make an agent designer sit up: it means the best move for every participant is to reveal its true preferences, because lying cannot improve its result. You get a market with no currency, no ledger, no incentive to misreport, and an outcome no coalition can beat, all from a rule about following arrows.
The kidney exchange is this algorithm wearing surgical scrubs, with one real-world constraint bolted on. Because every operation in a swap has to happen almost simultaneously, so that no donor can back out after their partner's patient has already received an organ, the cycles have to be short. A two-way or three-way swap is routine; a twelve-way simultaneous cycle is a logistical nightmare. So kidney exchange is Top Trading Cycles with a cap on cycle length, plus a lovely extension: a single altruistic donor, someone who shows up wanting to give a kidney to no one in particular, can kick off a chain rather than a closed loop, and a chain does not need to be simultaneous because no one at the front is waiting to be paid back. The technology that replaces money in this market is not a better coin. It is a matching algorithm.
There is a tempting wrong answer to "market without money," and naming why it fails is what sharpens the whole idea. The intuitive moneyless economy is the gift economy, made famous by the anthropologist Marcel Mauss in his 1925 essay The Gift. Mauss studied societies where valuables circulated not through sale but through giving, and he found that gift exchange rests on three obligations: the obligation to give, the obligation to receive, and, crucially, the obligation to reciprocate. A gift must eventually be returned.
That last obligation is the catch, and it disqualifies the gift economy from our brief entirely. An unreturned gift is a debt. The memory of who has given what to whom, and who therefore owes whom, is an account. A gift economy runs on a ledger, just one written in social memory and reputation rather than in a database. It is arguably the oldest ledger there is. So if your constraint is truly no ledger of any kind, the warm, humane, reciprocal gift economy is out, because it stores obligation across time, and stored obligation is exactly the thing a ledger is for. What satisfies the constraint is stricter and colder: a market where nothing is ever owed because every exchange clears completely in the moment it happens. The moneyless market the brief actually asks for is not the gift. It is the cycle.
Now bring it to agents, because this is where a piece of Nobel economics turns into a security decision. When you are designing a marketplace where software agents trade with each other, the reflex is to mint a medium of exchange: a token, a credit balance, a settlement ledger. And the moment you do, you have built an attack surface, three of them, actually, each of which the matching-market design simply does not have.
There is no token to inflate or corner, because there is no unit of account whose supply anyone can manipulate. There is no ledger to falsify or reorder, and that second one is bigger than it sounds. We have written before that MEV, the extractable value that miners and validators skim by reordering and front-running transactions, is coming to the agent marketplace. MEV is, definitionally, the exploitation of control over the ordering of a ledger. A market that clears in atomic mutual-want cycles has no ordering to exploit. There is no pending-transaction pool to watch, no sequence to rearrange, no sandwich to slip around someone's trade, because a cycle either executes whole or does not execute at all. Remove the ledger and you do not mitigate MEV; you delete the category. And there is no stored credit, so no agent is ever left holding an IOU that a counterparty later defaults on, because there are no IOUs. The absence of the medium is not a missing feature. It is the security model.
I want to be honest about the bill, though, because a moneyless market is a real tradeoff and pretending otherwise is how you get burned. You can only trade when a cycle exists. The double coincidence of wants that money was invented to escape comes back as a multi-coincidence, and it comes back harder: thin markets, where the right loop of wants does not happen to close, simply do not clear. Money buys liquidity, the ability to transact even when no direct match exists, and a moneyless market gives that up. You also give up storage of value across time; you cannot save, because there is nothing to hold. And the whole game moves from price discovery to preference revelation, which has its own strategic surface, which is exactly why the strategy-proofness of your matching rule stops being a nice-to-have and becomes the load-bearing property. A moneyless agent marketplace trades liquidity and savings for the elimination of tokens, ledgers, and every exploit that lives on them. That is a coherent trade, and for some markets it is the right one, but it is a trade.
So here is the practical residue, and it is more useful than "remove the money," which is a slogan, not a design. The real lesson is that a medium of exchange is a choice, not a law of nature, and it is a choice with a specific, nameable cost: every currency and every ledger you introduce is a surface someone can attack, inflate, reorder, or default on. Before you mint a token for your agents, ask whether the things they trade are matchable. If agents are swapping compute for storage, a dataset for a model call, an idle capability for a needed one, and if you can get them to reveal what they actually want, then you can clear those trades in cycles and simply never build the coin, never build the ledger, never inherit the MEV. You will pay for it in liquidity, and in thin moments the market will sit idle waiting for a loop to close, and you should decide up front whether that price is worth the attack surface you avoided.
The kidney exchange settles the question of whether such a market can exist, because it exists, at national scale, saving lives with no money and no memory of debt, running on nothing but a graph algorithm that finds loops of mutual want. The economists who worked it out won the highest prize their field gives, and the thing they built where the currency used to be was not a cleverer currency. It was a rule for following arrows until they close. If your agents can be made to point honestly at what they want, that same rule is waiting, and it comes with the one property no token has ever offered: nothing to steal, because nothing is ever stored.
A medium of exchange is a choice with a nameable cost. So is every other piece of trust machinery you bolt onto an agent market.
The discipline this essay asks for — before you mint the token, ask whether you actually need it, because it is an attack surface you will own forever — is the same discipline the agent trust stack is built around. It ships as separate installable pieces, not a monolith, precisely so you add only the trust machinery your market needs: verification against ground truth where an output has to be checked, a provenance record where the work has to be auditable, ratings where a track record has to be priced. Some markets clear in cycles and need none of it; some need one piece; few need all three. The point is the same as the kidney exchange's: decide what each mechanism costs you before you build it.
Read the Theory of Agent Trust
pip install agent-trust-stack · npm install agent-trust-stack
Or the pieces on their own: pip install chain-of-consciousness / npm install chain-of-consciousness.