What changed
Google made several agent-observability controls generally available across BigQuery and Looker: conversation logging into customer-owned tables, audit logging, BigQuery monitoring dashboards and Looker usage dashboards. Custom ADK or LangGraph agents can stream structured execution telemetry through an open-source plugin.
Why it matters
Teams can join agent traces, token use, latency, feedback and data-governance controls in infrastructure they already operate. That improves diagnosis and cost review while creating a sensitive telemetry store that must be scoped like production data.
What to do
Enable aggregate metrics first, then test conversation logging on synthetic prompts. Verify IAM, residency, CMEK, VPC Service Controls, row-policy fail-closed behavior, ingestion cost and deletion before adding real transcripts or tool arguments.
Caveat
Availability is mixed: core logging and dashboards are GA, feedback review remains preview and custom-agent instrumentation can capture fuller payloads. First-party traces intentionally omit some content, while conversation logs can contain prompts and responses and incur storage and query cost.