layman/terms
AI words

AI agent

ay · eye ay · jent

In shortAn AI that can take actions on its own, like searching, booking or editing files, to finish a task for you.

A chatbot tells you how to book a table. An agent books it.

Ask a friend how to post a parcel and they’ll explain. Ask a good assistant and they’ll take it to the post office, pay and bring you the receipt.

An AI agent is the second kind. It’s an AI that can take actions, one step after another, to finish a task you give it. It plans, does a step, checks the result and decides what to do next.

What agents can do

How they do it

The agent is an LLM that’s been given tools: a browser, a calculator, access to your email. It decides which tool to use and when. This is called tool use.

What it isn’t

It isn’t a robot. Agents are software. They also aren’t always right: they can misunderstand the task, take a wrong turn or get stuck.

Where the comparison breaks down

A human assistant knows when to double-check. Agents need clear limits, called guardrails, because they’ll keep going confidently even when something has gone wrong.

How the words fit together

An LLM that can use tools is doing tool use. Give it a goal and a loop and it becomes an AI agent. Guardrails keep it within safe limits.

Watch out

An agent acts with whatever access you give it. Start it on low-risk tasks. Ask it to check with you before spending money or sending anything.

When you'll hear it

“Let the agent handle it.” Give the task to an AI that can actually do the steps itself.

“It's agentic.” It doesn't just answer. It takes actions on its own to reach a goal.

“The agent got stuck in a loop.” It kept repeating the same steps without finishing.

Say it in a meeting

“What should an AI agent be allowed to do on its own? Where should it stop and ask a person?”