An AI becomes an agent when it can take actions in the world to pursue goals, not just produce outputs in response to prompts. The core shift is from "answers questions" to "does things."

The clearest marker is tool use — the AI can call functions, run code, search the web, control a browser, and send emails. But what really makes something an agent is the loop: it takes an action, observes what happened, updates its understanding, and decides what to do next. That perceive-reason-act cycle is what separates an agent from a one-shot response.