Google open-sources Ambient Quality Agent AQuA

Google has open-sourced AQuA, a reference implementation designed to diagnose production AI agent failures by analyzing recent sessions and comparing them against deployed source revisions. The tool processes up to 1,000 sessions through five stages to identify recurring failure patterns, verify candidate clusters, and propose code changes or attribute issues to external dependencies. In a controlled evaluation of a travel application, AQuA helped developers reduce specific failure rates through prompt modifications and session replays. Open article →