Classroom
Pick an exercise, duplicate a starter agent, and build. Each one ends with reflection questions so you think critically about what your agent did — and what it got wrong.
Exercise 1 — Build a fact-checking agent
✅ Fact CheckerCreate an agent that takes a claim and decides whether it is true, giving its reasoning and sources. Switch on the tools it needs and write a system prompt that forces it to never guess.
- • Which tools did your agent use, and why those?
- • Did it ever refuse to rate a claim? Was that the right call?
- • How could someone game your agent into a wrong verdict?
Exercise 2 — Build a citizen assistant
🧑⚖️ Citizen Support AgentUpload an official document (PDF or text) and build an agent that answers citizens' questions using only that document. Make it cite the source for every answer.
- • When you asked something not in the document, what happened?
- • Were the citations accurate? Click a retrieved chunk to check.
- • What's one risk of trusting this agent for real administrative advice?
Exercise 3 — Improve your prompt
Take any agent and save it as Version 1. Then change only the system prompt to make answers clearer or safer, and save Version 2. Use the diff and comparison views to show what improved.
- • What exactly did you change between versions?
- • Did the change help, hurt, or both? How do you know?
- • What would you test next to be more confident?
Exercise 4 — Detect media bias
📰 Media Bias DetectorBuild an agent that reads a news article from a URL and analyses its framing and bias even-handedly. Try it on articles from opposite ends of the spectrum.
- • Did the agent treat both articles with equal scrutiny?
- • Where did the URL Reader tool help, and where did it fall short?
- • Could the agent's own bias show up in its analysis?
Unlock to publish exercises for your students.