AI & engineering · 9 chapters · Free
Build an eval for an AI agent
An agent says it refunded a customer. The ledger says otherwise. Build the checks that tell the difference.
Start the courseWhat you’ll build
- A JavaScript grader that checks refunds against the ledger.
- A second grader that catches regressions and out-of-scope edits by a coding agent.
- An evaluation of two simulated workflows, with a decision you can explain.
Before you start
Basic JavaScript: functions, arrays and objects. Exercises run in your browser. No account, API key or paid tools.
Your code and first submissions stay in your browser. Export your work to keep a copy.
AGENT: “Refund complete!”
Requested: $24.00
Refunded: $0.00
Did the task actually succeed?
Inspect the evidence.
Write a check.
Find what your check missed.