"Agent" is one of the most used words in AI right now. Here is what it means without the hype.
Direct answer
An AI agent is a program built on an AI model that works toward a goal by planning steps, using tools such as search, files or code, checking the results and continuing until it finishes or gets stuck. Unlike a chatbot that only replies, an agent can take actions. People still need to set limits and review the outcome.
The agent loop
- Goal: "Find the three cheapest options and summarise them."
- Plan: decide the next step.
- Act: use a tool, such as a web search.
- Observe: read the result.
- Repeat or finish.
Chatbot vs agent
| Chatbot | Agent | |
|---|---|---|
| Output | Replies in text | Actions plus a result |
| Tools | Usually none | Search, code, files, apps |
| Steps | One response | Many, in a loop |
| Risk | Wrong text | Wrong text and wrong actions |
Examples
- Coding agents that edit files and run tests. See what AI coding agents are.
- Support agents that look up an order and draft a reply.
- Research agents that gather and summarise sources.
Building one
- Pick a narrow job.
- Choose a model.
- Give it a few safe tools.
- Set limits on what it can do.
- Test with cases where it should fail.
Risks
- Taking wrong actions
- Leaking data through tools
- Being tricked by text on a web page
- Costs from long loops
Keep a human in the loop for anything important.
How it fits with the rest
Agents sit on top of models. See how large language models work and how AI works.
Where Teyro fits
Teyro's The AI Stack course explains where agents sit among chips, models, data and apps, and you build your own AI assistant in your browser. It is free to start. To work with coding agents specifically, see Working with AI Coding Agents.
The bottom line
An agent is a model with tools and a loop. Powerful, and needs supervision.
