What an AI agent is (and how it differs from chat)

When people say "AI" today they usually mean chat: ask, read, close the tab. An agent is a different category — you give it a goal and it has tools to reach it.

Three things that set an agent apart

Its own environment. The agent runs in a container, not a chat window. It can store files, run commands and come back to unfinished work.

Memory. What it learns stays. You do not re-explain who you are every morning.

Work without you. An agent can run on a schedule — read the web, build a digest, save it. You just read the result.

A concrete example

Prompt: "Go through competitor pricing and build a table."

A chatbot tells you how to do it. An agent opens the sites, copies the prices, builds the table, saves it as a file and hands you a link. Next week it can do it again with the same prompt.

When an agent is overkill

For a one-off answer to a simple question. Chat is faster and cheaper there. An agent pays off when work repeats or when files and systems are involved.

What it means for price

Chat is billed per message. An agent is billed for the environment it runs in — which is why sleeping when idle matters. More in What an AI agent costs.