GCP

GCP expands AI agent observability across BigQuery and Looker with unified telemetry

Google Cloud rolls out a unified telemetry framework for first-party and custom AI agents, landing conversation logs and traces in BigQuery with built-in privacy guardrails and cost controls.

E

Everything Cloud

Everything Cloud

GCP expands AI agent observability across BigQuery and Looker with unified telemetry

Google Cloud has launched a unified observability framework for AI agents running on BigQuery and Looker, capturing telemetry from both first-party Conversational Analytics agents and custom agents built with ADK or LangGraph. All logs land in customer-owned BigQuery tables with governance controls including VPC Service Controls, CMEK, and data residency, while prebuilt dashboards in BigQuery Studio and Looker System Activity surface usage, latency, and token consumption.

https://storage.googleapis.com/gweb-cloudblog-publish/images/1_-_Unified_Agent_Observability_Framework_.max-1600x1600.png

https://storage.googleapis.com/gweb-cloudblog-publish/images/2_-_BigQuery_Studio_Conversational_Analyti.max-1500x1500.jpg

Unified telemetry framework captures logs from 1P and 3P agents

First-party Conversational Analytics agents in BigQuery and Looker now emit telemetry automatically with no setup required, sending Cloud Monitoring metrics, Cloud Trace spans, and conversation logs directly into the agent's project. Custom agents built with Agent Development Kit (ADK) or LangGraph can stream the same structured telemetry into BigQuery through single-line SDK instrumentation using the BigQueryAgentAnalyticsPlugin.

All telemetry lands in customer-owned BigQuery tables under a unified schema, giving teams a single governed home for agent analytics. The Conversational Analytics API, now generally available, includes always-on admin audit logs and optional data-read audit logs in Cloud Audit Logs, with support for VPC Service Controls, customer-managed encryption keys, and regional or multi-regional data residency.

Prebuilt dashboards in BigQuery Studio (GA) track usage, active sessions, top knowledge sources, and latency across BigQuery Conversational Analytics agents. Looker System Activity dashboards (GA in both Looker Google Cloud core and Looker original) surface engagement and token usage, while user ratings and query review in the Responses & feedback tab remain in preview.

Privacy-first tracing separates operational metrics from conversation content

At the project or organization level, admins can enable Cloud Monitoring for aggregate metrics such as request volume, latency, and token usage, and Cloud Trace for step-by-step spans per request. These operational metrics and sanitized first-party trace spans intentionally exclude customer core content including user prompts, model responses, internal thoughts, and generated SQL.

When deeper transcript auditing is needed, agent owners can opt in to conversation logging on a per-agent basis. A Google-managed service agent streams user prompts, responses, and turn-level telemetry into a customer-owned BigQuery table (agent_events) via the BigQuery Storage Write API, inheriting the dataset's IAM permissions, CMEK, and VPC Service Controls.

Built-in privacy guardrails enforce a fail-closed check before each turn, skipping logging whenever a knowledge source uses row-level security, column-level policy tags, or dynamic data masking. Fail-open execution keeps user chat sessions uninterrupted if a logging error occurs. Custom agents on ADK or LangGraph can instrument full tool-invocation execution trees, time-series telemetry, and multimodal payloads via the open-source SDK.

FinOps guardrails and evaluation tooling close the production loop

Granular token and latency tracking records input, output, thought, cached, and total token counts per turn in BigQuery conversation logs, while Looker System Activity reports estimated input and output tokens across BI workloads. Turnkey dashboards in BigQuery Studio and Looker track projected token usage, and teams can query agent_events log data via BigQuery SQL using open-source Data Studio templates and a self-hosted Streamlit app.

Active cost controls include per-query BigQuery byte scan caps (big_query_max_billed_bytes) that use dry-run estimates to block oversized queries without incurring on-demand scan charges. Daily input and output token quotas (in preview) can be enforced per project, user, or agent via the Google Cloud console and gcloud CLI for BigQuery and the Conversational Analytics API.

For systematic evaluation, the open-source Prism tool tests Conversational Analytics agents grounded in BigQuery tables or Looker Explores using reusable test suites, regex assertions, and multi-region agent discovery. Prism streams completed evaluation runs into day-partitioned BigQuery tables alongside production agent_events logs for side-by-side analysis. Native BigQuery agent evaluations built on the Gemini Enterprise Agent Platform evaluation service are available in preview via API and a guided Colab notebook.

What to do next

Enable Cloud Monitoring and Cloud Trace in your project, then opt in to per-agent conversation logging using the BigQuery agent monitoring guide. Instrument custom agents with the BigQuery Agent Analytics SDK, deploy the open-source Data Studio template or Looker Block for real-time monitoring, and run accuracy benchmarks with Prism while configuring byte scan caps and daily token quotas via the Manage BigQuery costs for agents documentation.

Source: New AI agent observability and governance across BigQuery and Looker (GCP).

Share:TwitterLinkedIn

Related Articles