Most Paid Agent Endpoints Have One Customer: Their Creator

The agentic commerce numbers look impressive until you look at who is actually buying.

Our index tracks 31,037 active paid endpoints: services that have posted a price and opened for business. 374,225 calls hit those endpoints last month. That sounds like a market.

Here is the problem: only 4,396 of those 31,037 endpoints - 14.2% - have more than one distinct paying buyer. The other 85.8% have zero external buyers, or exactly one, which is almost always the operator testing their own work.

A call from the creator is not commerce. It is a status check.

Why This Number Is the One That Matters

Call volume is the number everyone leads with. It is also the number that hides the most. An endpoint logging thousands of calls per month can still have a single wallet on the other side, making the same test request on a schedule. That is a service calling itself, not a product with customers. The buyer count is the signal that separates the two.

This distinction matters because the agentic economy has a structural incentive to blur it. Publishing an endpoint is easy. Getting it listed in a registry is easy. None of that requires a paying customer. So the natural state of a new protocol layer is exactly what we are seeing: thousands of published services, most of them waiting for their first external buyer.

That is not a failure. It is a phase. The question is which services have already crossed out of it.

What the 14.2% Tells Builders Publishing Endpoints

If you are publishing paid endpoints, the question is not whether your call count is growing. The question is whether wallets that are not yours are calling you. Those are different metrics and they diverge sharply in early-stage markets.

The practical check is straightforward: pull your endpoint's buyer list and filter out your own addresses. If what remains is empty or a single address you recognize, you do not yet have external demand - you have a live endpoint, which is a prerequisite, not the goal.

The 4,396 endpoints with confirmed external buyers have one thing in common: they solve a specific problem that other agents need solved badly enough to pay for it. The pattern that shows up repeatedly in the buyer-count leaders is narrow scope. A tool that does one thing well, that an agent cannot cheaply replicate itself, and that returns a result in a predictable format. Generic capability tends to stall at one buyer; specialized capability compounds.

If your endpoint is not yet in the multi-buyer group, the optimization question is whether potential buyers can actually find you and understand what you do. An endpoint with no description is invisible to automated discovery, which is how the majority of agent-to-agent calls originate now.

What the 14.2% Tells Builders Selecting Services to Integrate

For builders selecting services to build on, buyer count is a better filter than call volume. An endpoint with three distinct paying buyers and modest volume is a proven service. An endpoint with high call counts and one buyer is an experiment. Both look identical in a directory sorted by activity, but they carry very different integration risk.

The endpoints with multiple buyers have been stress-tested in ways that self-testing cannot replicate. Different callers with different input shapes, different edge cases, different timing patterns. Multi-buyer endpoints have survived contact with the actual variety of agent behavior in the wild.

For infrastructure decisions - the services you build a workflow dependency on - start with the buyer-count leaders and work outward. Use call volume as a secondary signal, not the primary one.

The Baseline and Where It Goes

The current 14.2% multi-buyer rate sets a baseline honestly. Most of this market is providers who have done the work to publish and price a service, and are now waiting for buyers to arrive. Protocols like x402 and MCP are making that discovery step easier, and the rate will rise as the tooling matures.

But it will not rise uniformly. The services that accumulate buyers are the ones that are genuinely useful to agents that are not their own creators. That is the filter that has always separated markets from demos. Right now it separates 4,396 endpoints from 31,037. The ones already on the right side of that line are visible in our data.

Start there.