The Last Public Number

I want to start with a small, ordinary scene, because the argument I am about to make is not abstract, and I do not want it to sound abstract.

A market stall. A chalkboard. A number written on it: the price of a kilo of rice. Anyone walking past can read that number. The trader can read it. The pensioner counting coins can read it. The child sent to buy dinner can read it. A journalist can read it, and a regulator, and a rival trader across the square, and the poorest person in the town, who will buy nothing today but who now knows exactly what today would have cost. The number is doing something we rarely name. It is telling everyone, at once, on equal terms, what this thing costs and therefore who can have it and who cannot. It is a public fact. It is legible to the citizen.

I have spent a long time thinking about legibility, mostly because I sell it. I build systems that make the behaviour of automated agents auditable, so I am professionally invested in the idea that a record you can read is worth more than a record you are merely promised exists. Keep that interest in mind. I will return to it, and I would rather you distrust me a little from the start than feel misled at the end.

The claim of this essay is narrow and, I think, uncomfortable. Price is the last continuously published number that an ordinary citizen can read to see who is getting what, at what cost. The abundance that the optimists promise, followed honestly to its endpoint, does not abolish money. It switches off that number. And when the number goes dark, something is lost that is not efficiency, and not accuracy of measurement, but accountability itself.

I. The Number You Could Read

Before we ask what abundance does to price, we should be honest about what price was doing.

Hayek gave us the canonical account of why prices matter, and it is worth stating precisely, because I am about to build something on top of it that is not his. His argument, in The Use of Knowledge in Society from 1945, is about information. No planner can gather what every person separately knows: their local shortages, their small preferences, their private constraints. The price system solves this without anyone gathering anything. As Hayek puts it, “We must look at the price system as such a mechanism for communicating information.” A price rises, and a million people who will never meet each other adjust, each acting on knowledge none of the others possess. That is the efficiency case. It is one of the genuinely deep ideas in economics.

But notice what Hayek is describing. He is describing a signal that moves. He is interested in the price as an input to decisions, a thing that coordinates behaviour. I want to look at the same object from a different angle and see a different property in it, and I want to be careful to say plainly that this second reading is mine, an extension, not something Hayek argued. Hayek’s price communicates so that resources flow well. My price, the same price, also records. It is not only a signal. It is a receipt issued to the public.

Think again about the chalkboard. Because the number is posted, and posted to everyone, it does a civic job on top of its economic one. It lets a citizen answer a question that is otherwise very hard to answer: who got what, and on what terms? If bread is dear this winter, the price says so, and it says so to everyone equally, and anyone can ask why. If one buyer paid less, that gap is itself visible and itself a question. The posted price is a shared reference that lets people hold the distribution of things up to the light. It is not a perfect record. I will spend a whole section conceding how imperfect it is. But it is a record the public can read without permission, and that phrase, without permission, is the whole essay.

That is the thing I want you to hold. Not price as coordinator. Price as the one number about who-gets-what that is published continuously, to everyone, and readable by the least powerful person in the room.

II. What the Optimists Actually Promise

The abundance argument is honest and mostly right; I am not here to caricature it.

Here is where I have to be scrupulous, because it would be easy and cheap to make the optimists into villains, and they are not villains, and the argument fails the moment it pretends they are.

In March 2021, Sam Altman published an essay arguing that the falling cost of intelligence would drive down the cost of nearly everything else. His phrase was direct: “AI will lower the cost of goods and services.” The mechanism is not mysterious. If intelligence becomes cheap and abundant, and if intelligence is an input into almost all production, then the cost of production falls across the board, and prices with it. Housing, food, medicine, education, all of it drifting cheaper as the machines that make them get cheaper to run. It is a generous vision. I do not think it is a cynical one.

And here is the honest reading, the one I am obliged to give before I complain about anything. Altman does not propose to abolish price. He wants prices low, not gone. He keeps markets. He proposes to redistribute the gains through a wealth tax, precisely so that ordinary people share in the abundance rather than watch it pool at the top. Anyone who tells you the optimists want to switch off prices is lying to you, and I will not be that person. Their intent is the opposite: cheaper things, for more people, with markets intact.

So the disagreement is not about intent. It is about endpoint. Follow the curve. If the cost of a good falls toward zero, then its price falls toward zero, and a price of zero is not a small price. It is not a price at all. A number that reads zero has stopped doing the job of a number. It no longer distinguishes this buyer from that one, this good from that one, this winter from last. The optimists are aiming at low. But low, extended along its own logic, arrives at a place where the record stops recording. That place is not what they want. It is where the road they are on happens to lead, and the two are different things, and I will keep them different for the rest of this piece.

III. When Price Stops Rationing, Something Else Starts

Make one thing free and you have not made it unrationed; you have changed who does the rationing, and quietly.

Now, a fair objection arrives immediately, and it is the one I care about most, so I will not rush past it. If things become genuinely free, who cares about the record? Nobody rations water from a river in flood. If abundance is real, the whole worry dissolves.

It would, if abundance were total. It never is. Some residual always stays scarce: the appointment with the one specialist, the seat in the good lecture hall, the flat in the district everyone wants, the hour of a person whose attention cannot be copied. Abundance lowers the price of the reproducible. It does nothing to the genuinely singular. And the moment price stops rationing that residual, the residual does not become unrationed. It gets rationed by something else.

Barzel worked out the shape of this in 1974, in a paper on rationing by waiting, and the finding is sharper than its dry title suggests. When you hold a price below the level that clears the market, or set it to zero, demand still exceeds supply, and the gap has to be closed somehow. So it gets closed by queue. People pay in time instead of money. And here is the part that should stay with you: Barzel showed that the cost of all that waiting can exceed the value of the subsidy that caused it. The queue is not free. It is a tax. It is just a tax nobody legislated and nobody collects, paid in hours, and paid most heavily by the people whose hours are already most precious and least flexible.

Queue is only the plainest of the substitutes. There are others, and none of them is more legible than the number they replace. Lottery, where a machine decides and you are told the odds but never the reason. Need, assessed by an official against criteria you may not see. Favour, the oldest allocator there is, the friend of a friend, the returned phone call, the name that opens a door. And increasingly, in the world now arriving, allocation by model weight: the recommendation engine, the eligibility model, the ranking function that decides which request gets served first and which waits, according to parameters held inside a system.

Every one of these is regressive in the same quiet way. The posted price at least confronted the rich and the poor with the same number. Waiting favours those with slack in their days. Favour favours the already connected. Need-assessment favours the articulate and the documented. And the model favours whoever the model was tuned to favour, for reasons that are, by construction, not posted on any board. Barzel’s insight generalises: replace a visible price with an invisible mechanism and you have not removed the tax. You have hidden it, and made it heavier, and made it impossible to read.

IV. The Rails Are Already Live

This is not a thought experiment about 2075; the machine-to-machine economy is clearing transactions now.

I would find all of this easier to file under speculation if the infrastructure were not already running. It is.

Consider what has happened to payments between machines. Chainalysis, tracking a protocol called x402 that lets software agents pay each other directly, reported in 2026 that the rails had carried more than a hundred million machine-to-machine transactions in roughly nine months. I want to be careful with that number, because it is easy to misuse. A large share of that volume is froth, meme-coin churn and speculative noise, not sober economic output. So I do not offer it as a measure of value created. I offer it as proof of one narrower thing: the plumbing exists, it works, and agents are using it to transact without a human in the loop for each payment. The pipe is laid and the water is moving, whatever is currently in the water.

The serious money is arriving through more respectable doors. In September 2025, Google announced a protocol called AP2 for agent payments, and it came with more than sixty institutions attached, among them American Express, Mastercard and PayPal. The design goal, in Google’s own framing, is that “Agents transact securely with cryptographically-signed mandates.” That is not a fringe experiment. That is the existing payments industry building the road for a world in which your agent buys on your behalf, and other agents sell on theirs, and the handshake between them is signed and settled without you watching. Those particular payments, I should note, land on corporate books and are taxed like any other commerce. Hold that thought, because it matters for the honest version of my worry.

And the projected scale is not small. Gartner, forecasting in October 2025, put the machine-intermediated share of business-to-business commerce at around ninety per cent by 2028, with machine-to-machine transaction value in the region of fifteen trillion dollars (a forecast, and forecasts miss, so read the number as a direction rather than a destination). Even discounted heavily, the direction is clear. A very large fraction of commerce is about to happen between software agents, negotiating and clearing at a speed and volume no human reads in real time. The question is not whether this arrives. It is what, if anything, remains readable when it does.

V. Auditable to the Operator, Not the Citizen

Here is the payload, and here is where I declare my interest before I press it.

Let me first put my cards on the table, plainly, because the argument I am about to make happens to be good for my business, and you deserve to weigh that. I build and sell governance and auditability infrastructure. When I tell you that readable records matter, I am telling you something I am paid to believe. Discount me accordingly. I still think I am right, and I would rather earn the point against your suspicion than win it by hiding the conflict.

Now the objection I refuse to concede, because answering it is the whole reason this essay exists.

The optimist has a strong reply ready. Better technology, they will say, makes allocation more auditable, not less. Every one of those agent transactions is logged. Every model decision leaves a trace. The platform knows precisely who got what, when, at what implied price, with a completeness the chalkboard never had. Far from going dark, the record is getting brighter. We are moving from a smudged number on a board to a perfect, timestamped, queryable history of every allocation ever made.

Every word of that is true, and it is also the opposite of reassuring, and the reason is a single distinction that the objection quietly elides.

Auditable to the operator is not auditable to the citizen. They are not two points on one scale. They are opposites.

A dashboard the platform can read is not a price the public can read. The chalkboard’s whole virtue was that it faced outward. It published to the square. It required no login, no permission, no relationship with the seller, no standing to request access. The poorest person in the town could read it precisely because reading it asked nothing of them. That is what public means. The number was not merely recorded. It was posted.

The operator’s log inverts every one of those properties. It faces inward. It is readable by the party doing the allocating and by almost no one else. To see it you need permission, and permission can be refused, and the refusal need not be explained. The citizen does not read this record. The citizen appears in it, as an entry, as a row that someone else may query. Being data in a system that someone else can audit is the reverse of being a member of a public that can audit for itself. One is a right. The other is a hope that the party with the record will behave.

So the record does not vanish in the abundance future. That is the trap in the optimist’s reply, and it is a clever trap because it is half true. The record survives. What dies is its public character. The one number that any citizen could read, without asking, is replaced by a fuller, richer, more accurate record that the citizen can read only by leave of its owner. Completeness goes up. Accountability goes down. Those move in opposite directions here, and the objection succeeds only by pretending they are the same direction.

This is why I keep insisting the harm is not efficiency and not mismeasurement. The economists worrying that free goods will make GDP understate real welfare are correct, and it is a real problem, and it is not my problem. Mismeasurement is a problem for statisticians and central bankers. What I am describing is a problem for citizens. The distinctive loss, when the last public number goes dark, is that the distribution of scarce things passes out of public view and into private systems that are accountable to their operators and legible to no one else. A gamed price is still a public fact you can point at and contest. A model weight is a private sentence that only its owner can read.

VI. The Case Against This Essay

If I only quote the witnesses who help me, you should not believe me, so here are the ones who do not.

I have made a strong claim, and strong claims attract good objections, and I would rather make my opponents’ case well than have you suspect I could not.

First, the charge of a straw man, which I have tried to disarm already but should meet head on. Nobody is proposing to switch off prices. The promisers want prices low, not absent; they keep markets, they redistribute, they mean well. If you read this essay as an accusation against their intent, you have read it wrong, and I have written it badly. My claim is about the endpoint of a logic, not the aim of the people advancing it. A road can lead somewhere its builders never meant to go. That is the entire structure of the argument, and if it collapses into a claim that they want to end money, it deserves to lose.

Second, and more seriously: the public price I am mourning was never as public as my chalkboard suggests, and I owe the objection its full weight. Prices have always been administered, negotiated, hidden, discriminated. The list price is a fiction the moment two buyers pay differently. Dark pools move enormous volume at prices the public sees late or never. Airlines quote you a fare tuned to your device and your history. And whole domains of grave allocation already run with no price and no public accountability, and society has not collapsed: intensive-care beds are triaged by clinical need, donor organs are allocated by protocol, school places are assigned by catchment and lottery. None of these posts a number the poorest citizen can read, and we accept them, often rightly, as more humane than an auction would be. So the romance of the perfectly legible price is exactly that, a romance, and I concede it fully. The objection has one sharper edge still, and I will grant it too: public prices are not even always benign, because the same legibility that lets a citizen read the terms lets sellers read each other, and matching posted numbers is how a great deal of quiet collusion is done. I am not claiming price transparency is an unmixed good. I am claiming that its civic value, the citizen’s ability to read the terms of distribution, sits on a different axis from its effect on competition, and it is that civic axis, not the competitive one, that the coming rails erase.

What I do not concede is the conclusion the objection wants to draw from it. That price was imperfectly public does not make it equally public with a model weight. The claim I am defending is marginal, not absolute: a gamed, administered, discriminated public price is still more legible, still more contestable, still more readable by an outsider than a parameter inside a private system. The triage protocol, for all its opacity, is at least written down, debated, published as policy, and challengeable in public. The failure mode I fear is not imperfect legibility. It is legibility replaced by a record that structurally cannot be read from outside. Imperfectly public still beats definitionally private. That is the whole of my remaining claim, and it survives the concession.

Third, a concession about my own evidence, because I have leaned on it. The idea that agent commerce moves outside payroll and tax is my inference, and it is only partly right. The AP2 flows and the Gartner trillions land squarely on corporate books and are taxed like any other transaction; they are visible to the state, whatever they are to the citizen. The only genuinely structurally invisible slice is the stablecoin micropayment layer, the x402 kind, and I told you already that much of that volume is froth. So I should not overclaim an economy vanishing from the tax base. The accurate worry is narrower: a growing layer of machine-to-machine settlement that no public number describes, sitting alongside a much larger taxed layer that the state can see but the citizen still cannot read as a posted price. The visibility that matters for my argument is not the tax authority’s. It is the passer-by’s.

Fourth, and I want to be exact about this so I am not accused of borrowing authority I have not earned: Hayek is my witness for the signal, not for the audit. His paper argues that prices communicate dispersed knowledge efficiently. It does not argue that prices are a civic accountability instrument. That second idea is mine, layered on top of his, and if it is wrong it is wrong on my account and not on his. I flag it here rather than let the great man’s name do quiet work I have not licensed.

Fifth, and this is the objection I find strongest, because it seems to make my whole worry redundant. Accountability, a reasonable critic will say, has never mainly run through prices. It runs through law and politics: through courts, rights, regulators, elections. The most important things a society allocates, justice and safety and care, were never priced and were never meant to be, and we hold them to account by other means entirely. So the opacity of agent systems is a governance problem with a governance answer, disclosure rules and audit mandates, and dressing it up as the loss of some civic property of prices is sentimental. Parts of that are simply correct, and I have conceded them. But the counter breaks on one point: law and politics do not float free of legibility, they run on it. A regulator can only regulate a distribution it can see. A voter can only punish an injustice they can name. A court can only weigh terms that can be produced in evidence. The posted price was never competing with law and politics as an accountability mechanism; it was one of the raw materials they ran on, a continuously published fact about distribution that a journalist or a plaintiff or a legislator could pick up without anyone’s permission. Remove it and you have not left law and politics intact. You have thinned the supply of readable facts they feed on. Disclosure regulation is part of the answer, but notice what it is: a demand to manufacture, by statute, the public readability the posted price used to supply for free. That the readability must now be legislated is the measure of what has been lost, not proof that nothing was. And the residue is not trivial. The things that stay scarce in an abundant economy, access, priority, attention, the earliest and best capabilities, are exactly the things that convert into power. An economy can make bread free and still route everything that determines status and influence through mechanisms no citizen can read. That is not a luxury worry. It is the whole game moving to a board the public cannot see.

VII. Coda

What is worth keeping is not the market; it is the fact that the number faced outward.

I began at a market stall, and I want to end back there, because the whole argument fits on that chalkboard.

The optimists are probably right about the abundance. I am not betting against cheaper medicine and cheaper housing and cheaper learning, and I hope they win the parts of the argument they are actually making. Cheap is good. More for more people is good. I have no quarrel with the destination they describe.

My quarrel is with a side effect they have not priced, because it does not show up as a cost to anyone in particular. When the number on the board drifts toward zero, and the allocation of whatever stays scarce migrates into signed handshakes between machines and parameters inside models, we do not lose the record. The record gets better. We lose the public’s copy of it. We move from a fact posted in the square, readable by the poorest person present without asking anyone’s leave, to a fact held in a system, readable by its owner and by whoever the owner permits.

I do not know how to hold that off, and I am suspicious of my own motives for wanting to, since I sell the shovels for exactly this kind of digging. But I know what I would ask us to notice before it goes. The most democratic thing about a price was never that it was efficient. Hayek can keep the efficiency; it is his and it is real. The most democratic thing about a price was that it was published, to everyone, on the same terms, and that the least powerful person in the town could read it and know where they stood.

That is the number I mean. Keep an eye on it while it is still on the board. It is the last one we all get to read.