Ask a chatbot how to find a document and it may write instructions. An AI agent can be given a search tool, retrieve documents, inspect the results and choose its next action. Its defining practical feature is the ability to act and use feedback to make progress toward a goal.
A typical system combines a model, permitted tools, working context and stopping rules. Each action returns information, such as a file, a search result or an error. The next step can change accordingly. A predetermined workflow follows a programmed route; an agent has some freedom to select that route during execution.
Applications include research assistance, coding and customer support. Autonomy varies substantially. Reading a calendar and sending invitations are different permissions. When examining an agent, ask which tools it can access, what needs approval and how completion is verified. These details connect claims about intelligence to work the system can actually perform.