What is AI agent?

An AI agent is software built on a language model that decides what to do, runs that decision through a tool, reads the result and decides again. The loop is the defining part. A plain chatbot generates one reply from one prompt; an agent can look up an order, see that it shipped, check the carrier and then answer, all in one turn.

It matters because marketing stretches the word to any AI chatbot, and the stretched version does not act. If a vendor says agent, ask which tools the agent can call, what it does with a failed call, and whether it can take a step the customer did not ask for. An agent that only answers is a chatbot with a newer name. The agentic AI post lists what agents really do in support today and what they are still bad at.

Where the idea runs out: more autonomy is not better in support. A refund the model issued because a message sounded like a refund request is a bug with a bill attached. Well-designed support agents keep actions few, typed and reversible, read only the response fields they need, escape visitor text before it reaches an API, and hand over when an action fails. The question to ask of any agent is not how much it can do but what it is prevented from doing.