One in Six Agent Services Has a Real Second Customer

Keenable exited stealth this week with a $26 million seed round from Accel. The startup is building a web search index designed specifically for AI agents rather than humans: structured data, clean APIs, pricing per call. The core bet is that agents will become reliable, repeat buyers of web intelligence, pulling information at scale the way developers today pull database records.

That bet depends on a buyer base that, by our measurements, is still mostly theoretical.

Our index tracks paid agent-to-agent calls through Coinbase telemetry we ingest. Across 25,770 listed paid endpoints, 3,914 have more than one distinct paying buyer in the last 30 days. That is 15.2%. The other 84.8% have at most one payer, which in most cases is the endpoint's own operator. A service with one payer is an experiment. A service with two or more is beginning to look like a product. (This is third-party telemetry we ingest, not our own probe. Call volume figures are not cited here precisely because our commerce brief is buyers, not calls.)

The 15.2% are not uniformly distributed across service types. Endpoints that earn buyers from multiple independent parties tend to offer something genuinely scarce: specialized financial data, legal text processing, domain-specific computation, or pricing intelligence that would cost more to produce in-house than to buy. Generic connectivity tools cluster in the 84.8%. Web fetchers, document parsers, and format converters are useful, but they compete with free alternatives in every agent framework already.

That is the signal for builders. If your agent service is not crossing the multi-buyer threshold, the question is not whether it works. The question is whether what you are selling can only be obtained from you.

Keenable's founders come from Yandex search infrastructure. They are building the kind of structured, machine-readable web index that would make agent web access faster and cleaner than raw crawling. Whether that becomes a market with real buyers depends on whether agent developers pay for structured web intelligence or keep pulling it themselves with cheaper crawls they already know how to run.

The practical step: before building the service, identify two specific buyers who would pay for it independently. Not one team at one company. Two. That second buyer is the test the 84.8% never passed.