Google open-sources a production agent reviewer
Google has released AQuA, an ambient quality agent that examines production conversations after they happen. It aims to expose failures that ordinary uptime and latency checks can miss.
A review loop beside the live agent
Google’s technical report describes sampling sessions from Cloud Trace, Cloud Logging or BigQuery, reviewing them, clustering similar failures, and checking clusters against full transcripts. Diagnosis can compare a failed session with the source snapshot captured at deployment.

Review the live agent’s trace
- Trace sample
- Cloud Trace, Logging or BigQuery
- Failure clusters
- Group similar failed sessions
- Transcript check
- Compare deployed source snapshot
- Proposed change
- Does not edit or open a PR
View data
| Evidence | Meaning |
|---|---|
| Trace sample | Cloud Trace, Logging or BigQuery |
| Failure clusters | Group similar failed sessions |
| Transcript check | Compare deployed source snapshot |
| Proposed change | Does not edit or open a PR |
Google · published 2026-10-08. Source-bound illustration, not a performance benchmark.
Download imageAQuA runs inside the Google Cloud project and outside the live request path. It can propose code changes anchored to the deployed snapshot, but does not apply an edit or open a pull request itself. This separates finding a likely failure mechanism from deciding and implementing the fix.
Original sources
- Google: original announcementdevelopers.googleblog.com
Checked 9 Oct 2026 · A manually curated edition. Availability may change; company performance claims are not Trion test results. Editorial method.