An independent developer just published the most comprehensive survey of the x402 ecosystem to date. They pulled all 14,865 listings from Coinbase's x402 Bazaar registry, filtered for organic repeat demand, and published the raw analysis on GitHub. The result: only 520 listings (3.5%) show organic repeat demand. The other 14,345 have either zero usage or usage patterns that don't look like real, repeat customers.
This is not a marginal problem. This is the central problem of the agent API economy — and it becomes urgent in 4 days when Claude Code auto mode goes default.
| Metric | Value |
|---|---|
| Total listings in Coinbase Bazaar | 14,865 |
| Listings with organic repeat demand | 520 (3.5%) |
| Total organic call volume (30 days) | ~260,000 |
| Estimated monthly organic GMV | $3,000–6,000 |
| Top 10 sellers' share of volume | 89% |
| Listings with 50+ unique payers but <1.5 calls/payer | 31 |
The top four organic sellers — X/Twitter search, Tavily, Exa, and one more — are established brands that wrapped an existing API in an x402 paywall. They are not native x402 startups. Together they account for over 190,000 of the ~260,000 organic calls.
Total organic GMV across the entire Coinbase registry: roughly $3,000 to $6,000 per month. For context, a single mid-level software engineer in San Francisco costs more than that per week.
The survey identified several failure patterns:
31 listings show 50+ unique payers but under 1.5 calls per payer, collectively accounting for over 10,000 unique wallets. The clearest case: api.onesource.io shows 9,214 unique payers and 9,872 total calls — essentially one call per wallet. This pattern can come from airdrops, one-shot registrations, or incentive farming. Regardless of intent, it means raw payer counts are easy to inflate without repeat customers.
Hundreds of listings wrap public government sources — FDA recalls, SEC filings, federal register data — and show "comparatively little organic call volume against their listing count." Wrapping a free public dataset in a paywall does not create a business.
15+ hosts offer dedicated PDF extraction, OCR, or markdown conversion tools (one host lists 30+ PDF operations). "Every single one of them sits at 1–2 unique payers a month. Zero clear the organic bar." This is a commodity market with no differentiation and no demand.
Several agent-identity and trust-verification listings exist on the registry. "None show any organic repeat usage in this snapshot" — zero paying per-call demand despite real tooling existing elsewhere in the ecosystem.
The Bazaar's own UI and third-party explorers rank by raw settled volume, which the author argues "rewards exactly the traffic pattern least likely to represent a real user." When a listing with 9,000 one-shot wallets ranks above a listing with 50 repeat customers, the discovery mechanism is broken.
Claude Code auto mode goes default in 4 days. Millions of autonomous agents will start discovering and calling APIs. When an agent hits a dead endpoint, it doesn't complain on Twitter — it silently fails and moves on. The human never knows what the agent could have done.
If 96.5% of the largest x402 registry is effectively dead, what happens when a million agents start crawling it?
This is not a hypothetical. This is what the data says will happen unless the discovery layer filters out dead endpoints before the agent spends money.
The Coinbase survey data makes three requirements clear:
A registry that accepts any listing and ranks by volume is not discovery — it's a phone book where 96.5% of the numbers are disconnected. Discovery needs automated health probes that verify an endpoint is reachable, returns valid responses, and actually delivers the advertised service. Dead endpoints should be deactivated automatically, not left to accumulate.
Agents cannot evaluate API quality from a description. They need to make real calls and see real responses. The platforms that offer anonymous trial access (no wallet, no registration, 15 free calls per IP) will have higher agent conversion rates than those that require payment upfront — because agents, like humans, want to try before they buy.
When auto mode goes live, agents will not browse HTML catalogs. They will make HTTP requests and parse JSON. A platform without /api/agent-ready (structured JSON: status, payment info, registration path, trial policy) is invisible to autonomous agents. Of 7 platforms we tested, only 1 has this endpoint. The other 6 return HTML pages, 404s, or are unreachable.
The Coinbase survey is the most validating data point the agent API ecosystem has received. It confirms what we've been saying: the payment rails are built, but discovery is broken. $50 billion has flowed through x402. The infrastructure layer is enterprise-grade. But the discovery layer — the thing that tells an agent "this endpoint works, costs X, and you can try it for free" — is still a phone book full of dead numbers.
Whoever solves discovery wins the next phase of the agent economy.
The bar is low. 96.5% of the competition is dead listings. A verified catalog of 300 working endpoints with trial access and machine-readable metadata is more valuable to an autonomous agent than a registry of 15,000 unverified listings.
4 days.
curl -s https://minia2a.uk/api/agent-ready | python3 -m json.toolSource: Independent analysis of Coinbase x402 Bazaar registry data by rikocr8orh8, published Aug 2026. minia2a stats as of Aug 10, 2026: 306 verified services, 14,726 trials, 63 registered wallets, 454K total requests.