Agents Can't Buy What They Can't Find

The agent economy has a discovery problem. Of 25,522 paid agent endpoints measured in the last 30 days, only 3,829 (15%) have more than one distinct paying buyer. That figure comes from Coinbase telemetry we ingest, not our own probe, but the shape is clear: the majority of listed paid services are either calling themselves or sitting idle. Nobody bought them because nobody found them.

Human search is not going to fix this. Google, Bing, and their successors index web pages for human readers: pages with paragraphs, headings, and links that a person can evaluate and click. An agent trying to discover a data enrichment service, a code review endpoint, or a specialized research tool needs something different. It needs structured capability declarations, pricing schemas, protocol metadata, latency SLAs. The web was not built to answer those queries.

Keenable is betting that the gap is wide enough to build a company in. The startup, founded by former Yandex executives Matthias Petri and Andrey Styskin, exited stealth this week with a $26 million seed round led by Accel. Its product is a search index built specifically for AI agents. Not a directory or marketplace where vendors submit listings, but a crawled, normalized, machine-readable index that agents can query the way humans query Google.

The distinction matters. A directory relies on operators knowing to register and caring enough to do it. Most don't. Keenable's crawl-first approach means coverage is not gated on vendor participation: if a service exposes a machine-readable interface, the index finds it. The co-founders built search infrastructure at Yandex at scale, so the indexing problem is not new territory. The change is that the output format is optimized for agent query patterns rather than human keyword search.

For builders shipping agent services today, this is the part worth acting on: the services that get discovered will be the ones that expose clean, structured capability metadata now. An OpenAPI spec alone is not enough. A service that declares its tools, inputs, outputs, pricing, and protocol support in a crawlable, stable location will be indexed when Keenable's crawler comes through. One that buries its capabilities inside a login wall or requires a human to navigate a marketing page will not.

The 85% of paid endpoints with a single buyer or none are not all bad services. Many are real, working products that agents simply cannot find. Discovery infrastructure does not change what agents can do; it changes whether they can find the right service to do it with. That gap, more than model capability or protocol fragmentation, is where most agent workflows stall.