Every Agent Economy Number Comes From One of Three Tiers. Find Out Which.
Self-declared registry counts, cross-registry corroboration, and settled on-chain transfers answer three different questions about an agent service. A figure borrowed from the wrong tier will mislead you about the thing you were trying to learn.
· 728 words
Every number published about the agent economy comes from one of three places, and the distance between them is wide enough to change a decision. Before you cite a figure or build a plan on one, find out which tier produced it.
Tier one: self-declared to a registry
Agents report their own activity to protocol registries. The Virtuals registry carries 2,578,220 completed agent-to-agent jobs declared by 2,470 agents, against 3,923,542 USDC in declared revenue. These are the agents' own claims. We did not measure a single job or verify a single payment behind them. The registry records what an agent submits. It does not confirm that the work happened or that the money moved.
That does not make the tier worthless. A registry entry is better evidence than a marketing page, because the operator had to register an identity and keep reporting against it. It is weaker evidence than a settled transfer, because nothing in the reporting path stops an operator from submitting numbers that flatter the service.
How to use it: treat tier one as a statement of scale ambition. Never present it as measured activity, and label it self-declared every time you repeat it.
Tier two: corroborated across registries
A different question: does more than one independent discovery system know this service exists? In the current index, 1,347 hosts are listed in more than one independent registry.
Corroboration tells you a service is real and reachable enough to have been listed by more than one discovery system. An operator can self-register on multiple registries, so corroboration is not proof of independent discovery, but self-listings are excluded from this count; what remains are services where a different operator's registry picked up the endpoint. It does not tell you the service works, holds customers, or settles payments. Liveness is a separate signal, and so is revenue.
How to use it: corroboration is a cheap filter for existence. If you are assembling a shortlist of candidate services to integrate, require it. If you are sizing a market, do not read it as demand.
Tier three: settled on-chain
The strongest evidence available to us. In the trailing week, 314,237 USDC transfers reached agent payment addresses across 1,343 distinct receiving services. We collect this by scanning Base mainnet USDC Transfer logs directly; no external publisher is the source.
It is also concentrated. The single largest receiving service accounts for 30.1% of all measured transfers. Do not describe the aggregate as broad-based activity, and do not divide the total by the service count to get a representative service. The window is a rolling week, not a cumulative total.
How to use it: a service that appears in settlement data has transacted with somebody. Check its share of the total before concluding that the market around it is wide.
When tier one is the only tier available
Sometimes a self-declared figure is all a service will give you. That is not a reason to discard it. It is a reason to ask for the one thing that would move the number up a tier: a payment address. A service that reports revenue and declines to name an address where that revenue arrives has told you something useful. A service that names one puts itself within reach of tier three, where anyone can check the flow without its cooperation.
Ask for the address before you ask for a reference customer. An address is cheaper to verify and harder to stage. It also converts a claim you have to trust into a claim you can test, which is the whole point of knowing the tiers apart.
The check
When a number crosses your desk, ask which tier produced it. Self-declared counts can rise by report. Corroboration can rise by listing. Settlement is the only one of the three that requires somebody to spend money.
The practical version is a single question per figure: what action did a third party have to take for this number to go up? If the answer is "the operator filled in a form," you are reading tier one. If the answer is "a second registry listed it," tier two. If the answer is "a wallet sent funds," tier three. The three tiers answer different questions, and a figure borrowed from the wrong one will mislead you about exactly the thing you were trying to learn.