Built a way for agents to walk through a product, capture what they saw, and hand reviewers useful release notes.
- role
- system designer / evaluator
- context
- Product QA and release review
evidence
- Turned a structured QA plan into browser walkthroughs
- Captured screenshots, traces, and findings for reviewer handoff
- Kept release judgment separate from agent execution
Autonomous QA becomes useful when it stops being a demo and starts helping a team decide whether a release is ready. The system needs to turn product intent into real browser behavior, capture evidence, classify issues, and leave a reviewer with enough context to make the call.
The SmartApply QA work is the clearest public case study: break a product surface into testable user journeys, let agents run the walkthroughs, and preserve screenshots, traces, and findings so the output survives handoff.
The goal is not to replace engineering judgment. It is to make repetitive exploration cheaper while making the system show what it saw, what it inferred, and what still needs human review.