A chatbot replies with text. An agent can also take steps: search the web, read files, call other software, and keep going until a task is done.
The agent loop
An agent works in a loop. It looks at the goal and the current situation, decides on an action, uses a tool, observes the result, and then repeats. The language model acts as the decision-maker, while tools do the actual work.
What tools add
With tools, a model can look up current information, run calculations, update a spreadsheet, or send a message. This extends what it can do well beyond text prediction.
Why agents raise the stakes
An answer that is wrong in a chat window is just text. An action that is wrong can delete a file, send an email, or spend money. Agents can also make small errors that compound over many steps.
Sensible safeguards
- Limit which tools and files the agent can reach.
- Require approval before irreversible actions.
- Keep a log of what the agent did.
- Start with low-risk tasks and expand gradually.
A useful mindset
Match autonomy to risk. Let agents handle routine, reversible work on their own, and keep a human in the loop where mistakes would be costly.
Continue with the book
Explore the full treatment, worked examples, and exercises in AI Made Simple.
Adapted from AI Made Simple, published by Choral Software. Article text and original examples © 2026 Choral Software. All rights reserved.