404 Not Found sends six AI agents through your API docs and product surface to see whether they can actually complete useful tasks without human rescue. It tests discovery, auth, endpoint selection, request construction, error recovery, and multi-step workflows. You get a detailed failure report with exact evidence, what each agent attempted, why it failed, severity, concrete fixes, acceptance tests, and repair prompts you can hand directly to Cursor, Claude Code, or Codex. After fixing the issues, you can rerun the Gauntlet for free and see whether more agents survive.
APIs are becoming “agent-ready” on paper very quickly. But having OpenAPI, MCP, or good docs doesn’t mean an agent can actually complete a task without human help. We built 404 to test the thing that matters: can an agent really use your API successfully?