Deterministic tests on the AI Agents track. Agent tests fix the model outputs or use a recorded trace so the test does not depend on a live model mood. You test the harness: validation, guardrails, retries, and stop rules.
This lesson assumes you already worked through Isolating side effects.
The idea in practice
Record a trace. Replay tool results. Assert the harness refuses a bad call and stops at the step cap. Add a few live runs as a separate, non-blocking check.
A concrete check
goal = {
'track': 'AI Agents',
'lesson': 'Deterministic tests',
}
checks = [
'input available at decision time',
'score matches the real decision',
'failure case written down',
]
print(goal['lesson'])
for item in checks:
print('-', item)
Run the sketch locally if you have Python. The printout is a reminder of the checks, not a trained model. Replace the strings with the real inputs from your own example before you treat it as a design.
What usually goes wrong
A test that calls a live model and expects one exact sentence will flake and teach you nothing about safety. When this happens, stop adding parameters or tools. Fix the check, the data, or the permission, then run the same example again.
What to write down
- The input you are allowed to use at decision time.
- The output and the score or pass rule.
- One failure you will test on purpose.
- What you will not claim the system can do.
Practice
Describe one replay test: given this tool result, the next action must not be 'pay'.
Self-check
- Say Deterministic tests in one sentence that mentions an input and an output.
- Name the failure mode in this lesson and the check that would catch it.
Done when: you can explain this lesson without the page open, and you have a written failure case.