AI agent
An AI agent is a system in which a language model plans and executes multi-step work toward a goal, using tools — web fetches, code execution, API calls — and feeding each result back into its next decision. Where a chatbot answers, an agent acts: it reads documentation, selects services, constructs requests, and recovers from errors with limited human supervision.
Agents are becoming a primary consumer of APIs, and the first mass wave is already observable: coding agents. When a developer asks Cursor, Claude Code, or Codex to integrate an API, the agent fetches the public documentation, decides what it can operate, and writes the integration — no sales call, no evaluation spreadsheet, just docs read by a model in seconds. Every scaffolded integration is an agent reading an API's docs in anger.
The mechanics of that reading are unforgiving. Most agents fetch over plain HTTP without executing JavaScript, navigate by machine-readable entry points like llms.txt and OpenAPI specs, and budget their context window across everything the task touches. What an agent can find and understand on that surface determines whether it chooses the API at all — which is the property producers can measure and improve.
Agent behavior is the frame for the whole Discry methodology: the discovery dimension checks whether an agent arriving on a plain fetch can find a machine-legible surface at all, and the comprehension dimension runs the reading as an experiment — real models quizzed on the live docs and graded mechanically against citation-verified ground truth.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free