A day in the life
- Review user sessions and feedback to spot where the AI confused people
- Prototype a new agent flow using a live model, not fake placeholder text
- Test edge cases like slow responses, refusals, and wrong answers
- Meet with engineers and researchers to review model behavior and evals (structured tests of AI output quality)
- Update designs for loading, citation, and undo states based on findings
- Document interaction rules and share open questions with the team
Tools you will use