AQuA runs alongside an agent in Google Cloud, outside its request path, and reads telemetry from Cloud Trace, Cloud Logging, or BigQuery. It samples up to 1,000 recent sessions per run and moves them through five stages: sampling, reviewing, clustering findings, verifying candidate clusters, and tracking issues. Developers can set domain-specific quality criteria in a goal.md file and add deterministic custom metrics, and the tool can run on a schedule, after a deployment, or on demand.
For root cause analysis, AQuA compares failing sessions with an immutable source snapshot captured at deployment. Its diagnosis agent cites file and line ranges and proposes changes when the cause lies in the repository, and it can also attribute failures to external dependencies, tool handoffs, or retrieved data. It does not apply fixes or open pull requests on its own.
In a test on 32 sessions from a multi-agent travel application, Google's verifier rejected three of nine candidate clusters, leaving six verified issues. After prompt changes and replays, seat-selection bypasses fell from 15 sessions to two and dietary-preference omissions dropped from seven to zero. Google notes these results came from controlled replay, not production traffic, and that sampling can miss rare failures.
Why it matters if you build agents
Teams running agents on Google Cloud can use AQuA to catch quality problems, such as skipped tool calls or ignored user preferences, that return successful responses and never trigger error alerts.
Findings link to supporting traces and must reference valid lines in the deployed snapshot, so developers can check each proposed root cause against the code before acting on it.
Replay may not reproduce external state, and sampling can miss rare failures, so the tool's findings should be verified rather than treated as complete.
Get stories like this every weekday
One short email with the day's agent tooling news.