AI agent

An AI agent is a system built on top of an LLM that goes beyond answering questions: it receives a goal, plans the steps, uses tools (web search, code execution, API calls, file reading) and evaluates results to decide the next step, repeating the loop until the task is done. The difference from a regular chatbot is autonomy: the agent performs actions in the world, not just text generation.

The typical building blocks are: an LLM as the "brain", a set of tools described to the model, memory (of what has been done) and a control loop. Frameworks like CrewAI and AutoGen make it easier to assemble agents, including teams of agents that collaborate, each with a role (researcher, writer, reviewer).

When to use it: multi-step tasks that require mid-course decisions: research with synthesis, process automation, support that queries systems. A concrete example: a post-sales agent receives "customer X requested a refund"; it looks up the order in the store's API, checks the applicable policy, issues the refund and writes the confirmation email, escalating to a human only if something falls outside the rules.