An AI agent is an AI system that works toward a goal by choosing each next step itself, such as running a search, calling a tool or writing code, based on the result of the previous step. Instead of producing one answer to one prompt, it loops: it acts, observes what happened and decides what to do next until it reaches the goal or stops.
This makes agents suited to open-ended tasks where the steps cannot be fixed in advance, such as researching a topic or working through a multi-step process across several tools.
The same flexibility creates risk. Errors can compound across steps, and an agent may take an action that was not intended, especially if it has broad permissions or reads untrusted content that steers it. Results can be less predictable and more costly than a fixed workflow.
To use agents responsibly, give them only the permissions the task needs, require human approval for consequential actions such as sending messages or changing data, log what they do and set limits on steps and spending. If a fixed sequence of steps would do the job, a simpler workflow is often more reliable.