Everyone Building Money for AI Agents Is Solving the Wrong Problem

Every few weeks another well-funded team ships another piece of the same solution to the same problem. One company raises sixteen million dollars to become the runtime that every agent's money flows through. Another raises eight to watch every transaction on the chain and stop the bad ones in time. A third, valued at a billion dollars, builds the universal network that agent payments are meant to route through, and holds none of the money while standing in the middle of all of it. A respected researcher publishes the case for hiding what the chain insists on revealing. And a builder who spent a year inside the space publishes the quiet conclusion that none of it has any customers yet. Each effort is sharp. Each is right about the single piece it holds. And the whole field is aimed at a problem that does not need to exist.

The problem they are all circling is how to make agent money safe when the system underneath enforces nothing on its own. So a layer grows on top: a managed runtime to route the spend, a monitor to watch it, a privacy service to hide it, a custodian to hold the keys. The work is real and the people are good at it. They are building an industry of trusted middlemen for a thing that was supposed to need none.

We built the thing that needs none. The rest of this is what that means, told through five of the people building middlemen.

The wrong problem, stated plainly

Here is the shared assumption nobody examines: that an agent cannot be trusted to spend on its own, so something must sit between the agent and its money and decide, in real time, what is allowed. Call that something a runtime, a screen, or a policy engine. It is a trusted intermediary, and the entire field is a race to build the best one.

The right question was never "who should sit in the middle." It was "why is anyone in the middle at all." Put the limit in the rail: a cap, a locked destination, an expiry, enforced by the chain at the instant a payment is formed, so a payment that breaks the rule cannot be constructed. Root the identity in a proof that reveals nothing rather than a database someone has to hold. Do that and the middle disappears. There is nothing to route, nothing to watch, nothing to hold, because the control already happened, inside the transaction, before it existed.

That is the rail. Here are three companies building the middle, and two people who can see it.

Foil one: the company that wants to be the runtime

In February, Sapiom raised $15.75M to, in their own words, give AI agents trusted access to the API economy. Their thesis is good and it is nearly ours: "payments cannot be an integration bolted on after the fact. They must be a primitive at the execution level." We agree with that sentence completely.

Then look at what they built to deliver it: "the managed runtime and dynamic routing that allows agents to operate safely across the machine economy," abstracting "identity (KYA), wallets, policy enforcement, risk controls, metering, billing and multi-rails settlement behind a single integration." They wrote that payment must be a primitive at the execution level, and then built a platform that every payment is bolted onto. The runtime is the bolt-on. It is the most well-funded version of the middle.

The execution level is the chain itself, not a runtime you route through. We put the same primitive Sapiom describes exactly where they say it belongs, in execution, by putting it on-chain, where no runtime sits between the agent and its money. Their identity is KYA, a managed record of who an agent is that they hold and you trust. Ours is personhood: a proof that a real, accountable human stands behind the spend, revealing nothing about who they are, held by no one. Their "verifiable context that strengthens controls as usage grows" is a database they own and you populate. Ours reveals nothing and is owned by no one. Same goal, opposite architecture: they are the trusted middle, and we removed it.

Foil two: the company that wants to watch the rail

In June, Range raised $8.3M to keep control of money once it moves at machine speed and cannot be reversed. Their writing states the premise perfectly: "a control that fires after settlement is not a control on a rail that settles instantly. It is a record of something you can no longer change." True, and it is our premise too. The disagreement is one question: where does the control sit?

Range puts it beside the rail, and to run it they have to see everything: every wallet, bank, custodian, and exchange, screened before each transaction broadcasts. We put it inside the rail, so a transaction that breaks the rule cannot be formed and there is nothing to screen. Range even measured why watching loses the race: in their own study, more than six thousand flagged addresses were already empty by the time anyone froze them, and when the Drift protocol lost $285 million in two and a half hours, the keys were never compromised. The monitoring around them simply could not stop a structured drain the mechanism still permitted. They built the finest log in the industry and raised eight million dollars to make it finer. It is the wrong problem, solved beautifully.

Foil three: the billion-dollar network that holds nothing and is still the middle

The strongest objection to everything so far is that some of these companies never touch your money, so where is the harm. Mesh is the answer, and it is the largest of all of them. In January it raised seventy-five million dollars at a billion-dollar valuation to be the universal network that agent payments route through, and this month it named itself the best autonomous agentic-payments platform. It is genuinely non-custodial: funds pass from accounts you already hold straight to the merchant, and Mesh keeps nothing. By its own accurate description, no one stands between you and your money.

So we read their code. Their entire public codebase is account connectors and an autogenerated API client. There is no on-chain program in it anywhere. It works the way Plaid works: you link an account you already have, and Mesh's server instructs that account to pay. The money never passes through Mesh, which is exactly how they can say they hold nothing. But the rule that decides how much, to whom, and until when does not live in the rail. It lives on Mesh's server and the connected exchange's interface, where it can be changed, frozen, or lifted by someone other than you. Non-custodial, and still the middle.

That is the cleanest proof of the whole argument. The problem was never only custody. It was that the control sits inside something you have to trust, a runtime or a monitor or a billion-dollar server, instead of inside the transaction itself. Take the custody away and keep the server, and you have relocated the middleman, not removed him. We put the rule in the rail, so there is no server to trust and nothing in the middle to begin with.

Witness one: the thinker describing the door

A few days before Range's raise, the researcher Billy Gao published a careful, correct essay arguing that what keeps the next trillion dollars off public chains is not speed but legitimacy and privacy. His synthesis is the sharpest statement of the destination anyone has written: you should be able to prove a fact without revealing the data behind it. He frames it as the future. He is right that it is the answer. It is also a future we already shipped: a payment on our rail proves it is authorized and within its limits without broadcasting the position behind it, and the party behind an account can be shown to be real and accountable without being named. Gao wrote an elegant description of a door. We walked through it some time ago. Credit to Gao: his essay is better written than most of our own internal notes on the subject, which is exactly why it is the tell. When the sharpest independent thinker in a field publishes the destination as a prophecy, and a company you have never heard of is already standing at it, the field is looking in the wrong direction.

Witness two: the builder who said it out loud

The most damning account comes from inside. A builder writing as jessy spent a year building agentic-payments infrastructure, talking to the teams shipping it at Stripe, Visa, Coinbase, and Google, and published the uncomfortable conclusion: "no real demand exists yet." The numbers are brutal. Coinbase reported sixty-nine thousand active agents and a hundred and sixty-five million transactions, while independent analysis put real volume at roughly seventeen thousand dollars a day, half of it test transactions. Stripe had a thousand enabled merchants and, at its own conference, single digits of agents actually transacting. And the line that should be on a Dexter slide: "Visa's agentic token requires as much as nine months of KYC and effectively a $250M revenue minimum to qualify. Only Amazon and Walmart-class companies can close the identity loop."

Then jessy names the disease directly: "conceptual blinders. When your business is payments, every problem looks like a payment problem. But payments are one piece of a bigger problem. The hard problem is coordinating work between agents and humans, verifying what was done, and settling outcomes. Coordination is the actual prize."

That is an insider, with no reason to flatter us, reaching the title of this essay on his own, and pointing at the exact ground we stand on. The identity loop that only closes for Amazon and Walmart is the loop our personhood proof opens to anyone. The trust and settlement that jessy calls the real prize is the bound, on-chain layer we already run. He spent a year to conclude the field is solving the wrong problem. We spent that year building the right one.

What the middle can never become

There is one more thing, and it is why the middle is not only unnecessary but unable to follow. When the rule lives in the rail and the limit holds no matter who operates the system, a lender can finally trust it. That unlocks what no middleman can offer: credit. An agent can spend money it does not yet have, bounded on-chain so the downside is fixed before the first dollar moves, and the work it does repays the draw. This runs on mainnet today.

A runtime can route that draw, a monitor can watch it, a network can settle it, and none of them can underwrite it, because none can promise a lender the limit will hold. Putting the control in the rail is what turns a payment system into a credit system. That is the ground they cannot reach, because the thing that makes credit possible is the one thing a middleman, by definition, gives up.

The pattern, and what it costs to ignore it

Line them up and the shape is undeniable. Sapiom is funding the best possible runtime to sit between an agent and its money. Range is funding the best possible monitor to watch a rail it does not control. Mesh raised to a billion-dollar valuation to be the network in the middle, and proved you can give up custody entirely and still be the middle. Gao is writing the most articulate case for hiding a ledger that should never have exposed so much. And jessy, who tried to build in this space for a year, says plainly that the field has its eyes on the wrong target. Three companies pouring capital into the middle, at sixteen million, eight million, and a billion-dollar valuation, and two sharp minds confirming the middle is the mistake.

It is the mistake. The moment the limit lives in the rail and a proof replaces the data, there is no runtime to route through, no ledger to watch, and no identity database to hold, because the dangerous thing was never possible and the private thing was never broadcast. The entire industry of trusted middlemen becomes optional, which is worse than wrong, because you cannot fund your way out of being unnecessary.

None of these people are foolish, and none of these problems are fake. Agents really cannot spend safely on rails that enforce nothing. Privacy really is the thing keeping serious money away. The identity loop really is closed to everyone but giants. The work is good. It is pointed at the symptoms of one decision nobody questioned: to build money that enforces nothing and reveals everything, and then spend a fortune building middlemen to make that survivable. We made a different decision. We built money that enforces its own rules and proves its own facts, so the middle is gone. The rest of the field is fighting hard, and fighting well, on a battlefield we left.