Google releases AQuA, an open-source agent that diagnoses production AI agent failures

Google has released AQuA (Ambient Quality Agent), an open-source tool that runs unattended in a customer's own Google Cloud project and samples production agent sessions from Cloud Trace, Cloud Logging, or BigQuery to find recurring failures.

Google's developer blog reports that in a 32-session sweep of a multi-agent travel concierge, AQuA verified six issues and traced two to specific prompt lines; after fixes, full-session passes rose from 5/32 to 13/32. Google notes that verification and diagnosis are model-based and that the tool only proposes edits.

According to Google, transcripts, source snapshots, and BigQuery tables stay inside the customer's project, and AQuA sits outside the request path without writing back to the agent. It never applies an edit or opens a pull request on its own. The release covers only the outer quality loop, offered as composable building blocks that Google says it will refine as it studies how teams run continuous agent quality.