Not every AI tool is an agent. A translation box on a website, the chat app you type into, and an assistant that books a trip all use a model, but they use it in different ways. These are the three ways you’ll meet most often.
Three Ways to Use a Model
Here’s how a single request, a chat, and an agent compare:
| What happens | What keeps it going | |
|---|---|---|
| A single request | One question goes to the model, and one answer comes back. Nothing else happens unless something acts on the answer. | Nothing. It’s finished after one answer. |
| A chat | Each message you send goes to the model along with the conversation so far, and the reply comes back to you. | You do, one message at a time. |
| An agent | The app sends the model what it needs, uses tools when the model asks, and checks with you when it should. | The app does, step after step, until the task is done. |
All three are built from the same thing. Each message in a chat is one request, and an agent sends many requests in a row, acting on each answer in between. What makes something an agent is the app around the model: it keeps the task going, offers tools, and acts on what the model asks for. That’s the agent loop at work.
