A regular chatbot answers one prompt at a time. An AI agent goes further: given a goal, it breaks the task into steps, decides what to do next, and uses tools such as web search, code, or apps to get there.
For example, told to research a topic and write a report, an agent might search the web, read several pages, take notes, and assemble a draft — looping until the job is done, with limited human input.
Agents are powerful but need guardrails, because they act rather than just suggest. That makes checking their work and limiting what they can touch especially important. They build on large language models for reasoning and prompt engineering to steer their behaviour.