Google releases AQuA, an ambient agent that diagnoses silent production failures in AI agents
Google introduced AQuA (Ambient Quality Agent), a tool that runs continuously beside production AI agents inside a Google Cloud project to find failures that don't show up in logs or health checks, such as skipped validation steps or dropped context between sub-agents. It samples production session traces, grades them against a checklist, clusters recurring failure patterns, verifies clusters against full transcripts, and tracks them over time as new, recurring, or resolved. On request, it can also trace a failure back to the exact line of code or prompt responsible, using an immutable source snapshot captured at deploy time.