What an AI Agent Actually Does

An AI agent is a system designed to pursue a goal, not just respond once and stop. A normal chat model answers the prompt in front of it. An agent goes further: it can break the task into steps, choose actions, call tools, inspect results, and decide what to do next. That makes it useful for real work like researching vendors, checking documents, summarizing findings, or preparing a draft based on multiple sources. The important idea is that the agent is acting with a goal in mind, even if the work takes several turns or tool calls. Good agents are built with constraints, tool access, fallback rules, and sometimes approval steps, so they stay useful without becoming unpredictable.