What an AI Agent actually is (and what it is not)

If you have used ChatGPT or a similar assistant, you already know what a model does: you type, it answers. An Agent is what you get when you give that model three extra things.

The three parts

1. A goal instead of a question. You do not ask "what is in this email?"; you say "every morning, tell me what I need to act on today." The Agent decides which smaller questions to ask on the way.

2. Tools. A tool is anything the model can call: read an inbox, search the web, write a row into a spreadsheet, send a message. Without tools a model can only talk. With tools it can do.

3. A loop. The Agent looks at the goal, picks a tool, looks at the result, and decides what to do next, until it believes the goal is met or it gives up. That loop is the part that feels like "it is working on it."

Put together: an Agent is a model with a goal, some tools, and permission to take several steps on its own.

What it is not

A test you can run in your head

Take any task you do weekly. Ask:

1. Could I write the goal in one sentence?
2. Do the inputs live somewhere a tool can read (inbox, calendar, a folder, a sheet)?
3. Would I be fine reviewing the result before it goes anywhere?

Three yeses means it is a good first Agent. Two yeses means it is a good second one. The projects here are all three-yes tasks.

What to do next

Read Prompt basics for work to learn how to write the goal, then Connecting tools safely before you give anything access to your accounts. If you would rather see a finished example first, open the Daily Work Email Briefing project and read only the outcome section.