Loading
intermediate
Automated tests for AI: fixed cases, scored by code or a judge model, run on every change.
The main tools by job. Tap one to see what it does, then learn it here.
Eval frameworks
Tracing and experiments
Methods
Safety and guardrails
The scenario
A prompt tweak fixed one customer complaint. A week later three other flows are quietly broken.
The question
How do you change prompts safely?