The Agent Economy Has Buyers. Not Many of Them.

TRM Labs published research last week tracking $52.7 million across 198.9 million settlements through the x402 agent payment protocol. Their finding: most of that money is not coming from AI agents. The measurement looked at who was spending. Our index looks at the other side - who is getting paid - and reaches the same conclusion from there.

Of the paid agent endpoints in our index, 4,772 have more than one distinct paying buyer in the last 30 days. The remainder have a single payer or none. One payer almost always means the operator calling their own service. An endpoint with one buyer is not a marketplace entry; it is a developer testing their own product.

The figure to track is not call volume - it is buyer count. Total calls to paid endpoints over 30 days are 427,045, which sounds like activity until you learn that most of those calls originate from a single wallet. A thousand calls from one sender is a loop, not a customer base.

What builders should do with this

If you are selecting agent endpoints to integrate, buyer count is the fastest proxy for demonstrated demand. An endpoint with ten distinct buyers over 30 days has already proven strangers will pay for it. An endpoint with one buyer has not.

The practical filter: look for platforms that surface distinct buyer counts alongside price and call volume. A service that only shows total calls gives you the number that is easiest to inflate. A service that shows distinct payers gives you the number that is hardest to fake. If a vendor does not publish buyer counts, ask directly. An honest vendor knows the number. A vendor who only talks about total calls is often telling you something.

Why concentration matters

The x402 settlement total that TRM examined is real money. The 427,045 calls our index measures are real calls. Neither number is wrong. What both hide is concentration.

In the USDC transfer data we track, one entity accounts for 29.4% of all measured transfers. The market has buyers; they are not distributed. That structure is normal at this stage of a marketplace - the earliest movers dominate before the long tail develops. The problem is that aggregate statistics describe a broad-based market while the underlying distribution describes a narrow one.

The self-declared layer

A separate dataset worth watching, and not conflating with the above: 2,470 agents across the Virtuals registry report completing 2,575,582 agent-to-agent jobs and $3,923,516 in revenue. These numbers are self-declared. The agents report them; no payment behind them has been independently verified. Registry-sourced data sits above a marketing claim and below an on-chain transfer in terms of reliability. Track it as directional signal.

Where the real market is

The honest picture of agent commerce right now is 4,772 endpoints with demonstrated second buyers, a median call price of $0.01, and a payment layer that is real but concentrated. Reporting the total endpoint count or total call volume without the multi-payer denominator produces an accurate aggregate and a misleading story about where demand actually lives.

The TRM research and our index disagree on nothing. Both say the same thing: the infrastructure exists, money is moving, and the buyers are far fewer than the headline numbers imply.