The State of Artificial Intelligence APIs

About this paper

Every company in this report sells the idea that machines should be able to do things on their own. So the rubric was pointed at them, asking the only question that matters: can a machine use them? 4,904 artificial-intelligence companies, scored the way an agent would read them — not the way a launch post reads.

The median company in the artificial-intelligence industry scores zero on agent-readiness. Band for band, this industry’s machine-readability is indistinguishable from the catalog average across every other sector tracked. The best-scoring AI company is a CRM vendor. The organization that authors the Model Context Protocol scores 4.3 on agent-readiness. And the 81.9% of this industry that took venture money publishes materially worse APIs than the 18.1% that did not. The full report has the resource taxonomy, the facet-by-facet Kin Score, the twelve agent dimensions, a four-item blueprint that is days of work rather than quarters, the fund-by-fund diligence table, and the reason none of this is regulation’s fault.

What's inside

1. Executive summary — can a machine use the AI industry?
2. The state of the industry — the biggest cohort, the average result
3. The resource taxonomy — what the industry actually exposes
4. The composite Kin Score — and what the zeros say
5. Agent-readiness — the industry's own product, measured
6. The blueprint
7. The standards layer — a protocol war on the last mile
8. The regulatory layer — the first regime has arrived, and it does not ask for an API
9. Standards and artifacts for the agentic turn
10. Security, auth, scopes and compliance
11. Provider-by-provider intelligence
12. The investable thesis
13. Where this is going

What you get for $500.00