State of ERC-8004 — Q3 2026
What is actually out there, and what actually gets paid.
This report measures two things about the ERC-8004 agent economy: which registered agents can actually be reached and used, and which paid x402 services are being paid by more than one buyer. It joins four dated data sets, so every figure can be traced to its source.
About this edition.
Publisher: Onchain Agent Intel Snapshot date: 2026-09-07. Every figure that moves day to day carries its own as-of date. Licence: Creative Commons Attribution 4.0 (CC BY 4.0): you may share and reuse it with credit. Format: A web page. Figures are frozen at publication and will not change silently. Future editions will carry their own snapshot dates.
Terms used in this report. ERC-8004 is an Ethereum standard that lets an AI agent register an identity on a blockchain, so others can look it up. x402 is a standard that lets software pay a web service per request, in the HTTP exchange itself, with no account or invoice. CDP is the Coinbase Developer Platform. Its public catalogue of x402 services is called the Bazaar. USDC is a dollar-pegged digital currency. Base and Ethereum are the two blockchains covered here.
§1 · What we measure, and why anyone reading this should care
This report is original measurement, not a roundup of other people's numbers. We join four independent sets of data on a single dated snapshot, using exact matching rules that we describe in plain words, so that every headline figure can be traced to its source and nothing is double-counted:
- The Coinbase CDP Bazaar — a snapshot of every web endpoint that CDP lists as payable over x402, with per-endpoint counts of calls and of distinct paying wallets over the last 30 days. Taken 2026-09-07 at 11:26 UTC, it reports 14,690 listed items across all the networks it covers.
- Our ERC-8004 identity index — our own database of every agent registration on the ERC-8004 registry contract (
0x8004A169FB4a3325136EB29fA0ceB6D2e539a432) on Base and Ethereum, as of 2026-09-07 at 14:19 UTC. It holds 22,916 Base rows and 2,827 Ethereum rows, belonging to 7,293 distinct on-chain owners. Each registration is a unique numbered token, so there are no duplicates by construction. - The ERC-8004 reputation ledger — the public feedback that clients leave about agents on the ERC-8004 reputation registry (
0x8004BAa17C55a88189AE136b182e5fdA19dE9b63) on Base, which we read into our own database. As of the snapshot it holds 29,128 per-agent summaries and 277,777 individual feedback events, all on Base. Ethereum feedback is not covered in this edition, so nothing here makes a cross-chain reputation claim. - A sampled link-resolution test — we picked 600 registrations at random on each chain and tried to open the metadata link each one points to (10-second limit, one request at a time per host). It ran on 2026-09-02 and is the only live web fetching behind this report. Nothing was re-fetched for the snapshot.
We state the matching rule and the snapshot date in full, so you can repeat the join yourself. We also state the cap we apply so that no single publisher dominates a count (50 endpoints per publisher). The cap makes the numbers smaller than the raw totals, so we say so.
The rest of this report is what those four sets say when joined on the questions that matter to a buyer: which agents can be reached and used, which are getting paid, and how those two groups overlap. The most useful finding is how those two groups overlap, and it is not where you would expect.
§3 · Sizing the market: a flat catalogue, a five-payer core, and a concentrated top
Over the roughly 24-day window covered by our daily snapshots (2026-08-15 to 2026-09-07), the Bazaar did not grow. Catalogue items moved from 15,056 on 2026-08-15 to 14,694 on 2026-09-07 (−2.40%), while distinct hosts grew from 1,546 to 1,897 (+22.7%) and distinct receiving wallets grew from 1,417 to 1,492 (+5.3%). Individual items come and go while the number of hosts and publishers drifts upward. It is churn, not accumulation.
Now look at what actually gets paid, on the same 2026-09-07 snapshot:
- 14,367 Base endpoints have at least one distinct paying wallet in the last 30 days (98.9% of the 14,424 Base endpoints in total). That filter barely restricts anything, because it admits endpoints with a single payer, where an operator paying its own service looks the same as a real customer.
- 757 endpoints (5.27% of the earning cohort) have five or more distinct paying wallets in 30 days, a rough bar for "at least a handful of different buyers". Ninety-five percent of the paid catalogue is below it.
- The top 10 endpoints by 30-day call count take 43.97% of the 297,904 calls made to the earning cohort. The top of the market is concentrated in a few services, and the rest is a long tail of endpoints with one or two payers each.
If you call the Bazaar "an emerging market", note that it did not grow by item count in this window, and its useful core (endpoints with five or more distinct paying wallets) is 757 endpoints. That is the pool in which "a buyer found this endpoint and paid" means more than one observation or an operator testing its own service. Use the five-payer number as the denominator when you quote adoption. The 15,000-item catalogue is not the market. It is only the list the market is drawn from.
Two caveats.
One: the CDP Bazaar is one x402 discovery source, weighted to Base, and it is a sample, not the whole population of x402 activity. Endpoints outside CDP's index (Solana, other chains, private catalogues) are not counted. Every figure in this section is limited to what CDP lists, and every fraction is a lower bound.
Two: an endpoint with exactly one paying wallet cannot be told apart from self-dealing using on-chain data alone. That is why the five-payer group, not the one-payer group, is the denominator to use.
Chainalysis, a blockchain analytics firm, published "Inside x402: 100M Agentic Payments on Base" in June 2026 (https://www.chainalysis.com/blog/x402-agentic-payments-adoption/). It looks at the same payment rail from a different angle, and it is independent evidence that the x402 rail is materially active. Do not read it as a count of agents: the article does not publish one. Both readings can be true at once: the rail is active, and the discoverable per-endpoint economy is 757 endpoints at the five-payer bar.