Every generation of technology promises to remove the middleman, and every generation ends up with new middlemen wearing new clothes. The web was supposed to disintermediate publishing and retail. It did, briefly, and then the aggregators arrived and took a larger cut than the publishers and retailers ever had. Web3 promised to disintermediate finance, and much of it re-intermediated within a few years through exchanges and custodians, because most people did not want to hold their own keys.
There is a lesson in the pattern, and it is not that disintermediation is impossible. It is that the middleman was never the product. Trust was the product. The middleman was just the least bad way to manufacture it.
Why the middleman exists
Strip any intermediary to its function and you find the same job. Two parties want to transact and cannot verify each other, so a third party stands between them and absorbs the uncertainty. The marketplace vouches for the seller. The escrow service holds the money. The card network eats the fraud. The platform adjudicates the dispute.
For providing this, the intermediary charges rent, and the rent is priced against the alternative, which is the transaction not happening at all. That is why intermediary margins are so durable. You are not paying for the matching or the hosting. Those are nearly free. You are paying for the confidence, and confidence has no substitute until someone builds one.
The rent has a second cost that shows up later. An intermediary trusted by both sides accumulates power over both sides. It sets the fees, owns the customer relationship, sees all the data, and writes the rules of its own court. What began as a service becomes a jurisdiction. Anyone who has negotiated with a dominant platform knows the feeling of appearing before a judge who is also your landlord and your competitor.
The referee does a different job
Now consider a different figure: the referee. A referee is present at the transaction but not between the parties. The referee does not hold the ball, take a percentage of the score, or choose who gets to play. The referee's entire function is to apply agreed rules to observable facts and declare the result, and the referee's entire value depends on having no stake in it.
This distinction, between standing in the flow and standing beside it, is what I mean when I say the agent economy needs referees, not middlemen. When two agents transact, something must confirm that the promised outcome was actually delivered before value changes hands. That is a refereeing job. It requires rules fixed in advance, facts checked against them, and a verdict neither side can bend. It does not require custody of the relationship, ownership of the market, or a percentage that scales with the stakes.
My years in enterprise security taught me the same distinction in a different vocabulary. The worst security architectures concentrated trust in one privileged component, and that component became the prize every attacker aimed at. The best architectures made trust unnecessary wherever they could: every claim checked at the point of use, every verdict reproducible from the evidence. Verify, don't trust is precisely the referee's posture. The referee does not ask to be believed. The referee shows the rulebook and points at the replay.
What honest disintermediation looks like
So here is my test for whether the agent economy is being built correctly, and it is a test I apply to my own work at Setix before anyone else's.
First, is verification replacing trust, or just relocating it? If confidence in a transaction still traces back to "because the platform says so," nothing has been disintermediated. The rent has a new collector. Verification counts only when the verdict can be checked by the parties themselves: the outcome definition was agreed before the work, the evidence is inspectable after it, and a disputed result can be re-derived rather than appealed to a throne.
Second, does the fee pay for a service or for a position? A referee earns per verdict, and the price reflects the cost of verifying. A middleman charges a percentage of the value passing through, which is a tax on position. When verification is honest, its price should behave like the price of any utility: falling with scale, indifferent to the size of the transaction it confirms.
Third, can the parties leave? The deepest intermediary rent is captivity: identities, reputations, and relationships that cannot move. In a refereed economy, an agent's verified record belongs to the agent. The proofs of past outcomes are portable facts, not loyalty points redeemable only in the company store.
The agent economy will be the largest trust-manufacturing problem ever attempted. Millions of counterparties, no human relationships, no time for courts. We can meet it the old way, by growing a new generation of landlords, or we can meet it with referees: neutral, inspectable, replaceable, paid for the whistle and not the stadium. I know which one I am building. The whistle is enough.