Google open-sources AQuA to diagnose production AI agent failures
Google has open-sourced AQuA (Ambient Quality Agent), a reference implementation that analyzes production AI agent sessions by sampling telemetry from Cloud Trace, Cloud Logging, or BigQuery, clustering recurring failures, verifying candidate clusters, and linking verified findings to the deployed source snapshot for root-cause hints. AQuA runs outside request paths, supports domain-specific goal.md rules and deterministic metrics, reports code/file ranges when repository changes are implicated, and does not auto-apply fixes; Google notes sampling and replay limitations.