How do you structure your test suite for non-deterministic AI outputs, and how do you approach regression testing for LLM agents at Clear?
