Google Cloud Puts Agent Telemetry in Customer-Owned BigQuery

By Daniel Okafor

Google Cloud expanded its AI agent monitoring and evaluation suite on November 13, 2026, adding customer-controlled logging, dashboards, privacy safeguards and spending limits across BigQuery and Looker. The update covers managed data agents and custom agents built with Agent Development Kit (ADK) or LangGraph.

Google Cloud said the release establishes a common telemetry path for first-party and custom agents. Managed Conversational Analytics agents emit telemetry automatically. Custom ADK or LangGraph agents can send structured telemetry through single-line SDK instrumentation. Google Cloud said the data lands in customer-owned BigQuery tables for analysis in its developer and administrator tools.

What is generally available?

Google Cloud designated two groups of core monitoring dashboards as generally available:

• BigQuery Studio’s prebuilt dashboards track usage, active sessions, top knowledge sources and latency.
• Looker System Activity dashboards for Conversational Analytics are generally available in Looker (Google Cloud core) and Looker (original). They cover engagement and token usage.

Managed conversation logging is also generally available, according to Google Cloud. User ratings and query review in the Responses & feedback tab remain in preview. Daily token quotas, native first-party BigQuery agent evaluations and some Looker feedback features also remain in preview.

How does Google Cloud separate operations from conversation content?

Google Cloud separated operational telemetry from conversation content so administrators and agent owners can govern the layers independently. Cloud Monitoring supplies aggregate metrics for request volume, latency and token usage. Cloud Trace provides step-by-step spans for each request. Its observability scopes can consolidate telemetry across multiple agent projects.

Google Cloud says “Operational traces deliberately exclude prompts and responses.” Its sanitized first-party trace spans also exclude customer core content: user prompts, model responses, internal thoughts and generated SQL. The separation lets platform teams monitor activity without automatically giving operational tools access to each interaction’s substance.

For deeper auditing, Google Cloud lets agent owners and editors opt into conversation logging for individual agents. A Google-managed service agent then streams prompts, responses and turn-level telemetry through the BigQuery Storage Write API into a customer-owned agent_events table.

The table inherits the dataset’s identity and access-management permissions, customer-managed encryption keys and VPC Service Controls. Google Cloud said those inherited controls tie detailed logs to the customer’s existing warehouse permissions and security controls.

Privacy guardrails check each turn before logging. Google Cloud said they skip a turn when a knowledge source uses row-level security, column-level policy tags or dynamic data masking. If a logging error occurs, the user’s chat session continues uninterrupted. The design separates broad operational visibility from narrower, explicitly enabled access to full interactions.

Google Cloud also said custom agents can retain more diagnostic detail through the open-source ADK plugin. The plugin can stream full tool-invocation trees, time-series telemetry and multimodal payloads into BigQuery. First-party conversation logging deliberately strips low-level tool arguments and internal SQL.

How do the spending limits work?

Google Cloud said each BigQuery conversation turn records input, output, thought, cached and total token counts. It also records total turn latency. Looker System Activity separately reports estimated input and output tokens across business-intelligence workloads.

Per-query caps on BigQuery bytes scanned use dry-run estimates. Google Cloud said the estimates block oversized queries before on-demand scan charges accrue. Separate daily input and output token quotas cover BigQuery and the Conversational Analytics API. The preview quotas can apply per project, user or agent through Google Cloud console or gcloud CLI.

The mechanism couples consumption reporting with active limits. Byte-scan caps address query spending before execution. Token quotas extend enforcement to agent consumption at project, user and agent levels.

How do evaluations sit beside production logs?

Google Cloud’s open-source Prism tool tests Conversational Analytics agents grounded in BigQuery tables or Looker Explores. Prism supports reusable test suites, regular-expression assertions and multi-region agent discovery. Its evaluations focus on SQL accuracy and answer quality before prompt, model or schema changes reach users.

Prism can write completed evaluation runs into BigQuery tables partitioned by day. The benchmark results sit beside production agent_events records for analysis in BigQuery and Looker. Native first-party BigQuery agent evaluations are separately available in preview through an API and a guided Colab notebook.

Google Cloud positions customer-owned BigQuery telemetry as the common layer for monitoring, cost controls and evaluation. That common layer also uses the customer’s existing warehouse permissions and security controls. For production teams, two design choices carry the most weight: where operational traces end and full conversations begin, and which controls are ready for daily use rather than preview.

For teams moving agents into production, the next test is operational: which preview controls become dependable enough to enforce at scale, and how detailed conversations remain separate from broad monitoring. The maturity of those controls will determine how much agent oversight can run through BigQuery without creating a parallel governance system.