AI Agent Statistics 2026

The core finding: most AI products can't be used by AI

The irony of the agent era is measurable. Of 2,832 AI products scanned, only 4% are fully agent-ready (score ≥70) — meaning an autonomous agent could discover the product, read machine-readable docs, authenticate, and actually operate it. 21% are partially ready, usually offering an API but no event delivery or delegated auth. The remaining 75% are, to an agent, just marketing websites.

Why it matters commercially: when a user tells an assistant "find me a transcription API and wire it up," the assistant can only choose from products it can operate. Agent-readiness is quietly becoming a distribution channel — and the gap between the ready minority and everyone else is a real, quantifiable moat.

Adoption by capability

Capability	Products	Share of scanned	What it enables
Public API	610	22%	Agents can call the product programmatically
MCP server	73	3%	Native connection from Claude & MCP clients
Webhooks	97	3%	Agents subscribe to events instead of polling
OAuth	146	5%	Users can safely delegate access to an agent

The ordering tells a story. APIs are the mature layer — 22% coverage reflects a decade of API-first culture. MCP at 3% is the one to watch: the protocol only emerged in late 2024, so this is near-vertical adoption, concentrated in devtools and infrastructure. Webhooks and OAuth lag because they're the hardest to ship — and that's precisely why they carry defensive value for the companies that have them.

Readiness by category

Average agent-readiness score for every category with 20+ scanned products:

Developer-facing categories dominate the top — their buyers demand programmatic access, so agent-readiness came free. The bottom of the table is the opportunity: vertical AI categories where products are agent-operable in principle but haven't shipped the plumbing. A vertical product that ships MCP + OAuth before its competitors effectively becomes the only option an agent can recommend and use in its niche.

The agent-builder boom

The supply side is expanding as fast as the readiness gap: 460 companies in the directory build AI agents as their product — the largest single category — spanning coding agents, browser agents, AI SDRs, support agents, and the eval/observability layer around them. A grounding note from the demand side: even the biggest platform operators describe agent progress as slower than hoped — Meta's CEO told staff exactly that in July 2026 — which matches our data. The plumbing (this page's subject) is a prerequisite the ecosystem is still laying.

Methodology, briefly

Every listed company is probed automatically: docs pages, /openapi.json and /.well-known/ endpoints, llms.txt, robots.txt AI-crawler policy, and homepage claims that are then verified. Signals are weighted (API 30, MCP 25, webhooks 15, OAuth 10, docs 10, SDKs 10) into the 0–100 score shown on every profile. Full reference: agent-readiness docs. Test any site yourself with the free checker, or browse the agent-ready leaderboard.