Google Cloud adds unified observability for BigQuery and Looker AI agents

Google Cloud is expanding AI agent observability and evaluation across its Data Cloud, giving teams a shared way to monitor first-party agents in BigQuery and Looker alongside custom agents built with ADK or LangGraph. The company says the tools are designed to help teams trace agent activity, protect sensitive data, control costs and test answer quality.
- Google Cloud is unifying telemetry for first-party and custom AI agents.
- Conversation logs can be stored in customer-owned BigQuery tables.
- BigQuery Studio and Looker provide agent monitoring dashboards.
- Token quotas and query byte caps help control agent spending.
- Prism supports reusable evaluation tests before production changes.
Telemetry lands in customer-owned BigQuery tables
Google Cloud is rolling out a unified telemetry framework for its agentic data products. The framework captures information from first-party and third-party agents, writes it to BigQuery tables controlled by the customer, and makes it available for analysis in existing developer and administrator tools.
Conversational Analytics agents emit telemetry without additional setup, according to Google Cloud. Queries routed to BigQuery data agents in Gemini Enterprise through agent-to-agent workflows, which the company identifies as generally available, send Cloud Monitoring metrics, Cloud Trace spans and BigQuery conversation logs to the agent's project. Custom agents built with Agent Development Kit or LangGraph can send structured telemetry through single-line SDK instrumentation.
Managed conversation logging in BigQuery, also identified as generally available, lets customers inspect Conversational Analytics traces, including prompts and responses, in tables they own. The Conversational Analytics API includes always-on administrator audit logs and optional data-read audit logs in Cloud Audit Logs. It also supports VPC Service Controls, customer-managed encryption keys and regional and multi-regional data residency.
Monitoring, privacy and cost controls
BigQuery Studio provides prebuilt dashboards for usage, active sessions, knowledge sources and latency across BigQuery Conversational Analytics agents. In Looker, System Activity dashboards track engagement and token usage for Conversational Analytics. User ratings and query review in the Responses & feedback tab are listed as being in preview.
Google Cloud separates operational telemetry from conversation content. Administrators can enable aggregate Cloud Monitoring metrics, such as request volume, latency and token usage, as well as step-by-step Cloud Trace spans. The company says sanitized first-party trace spans and operational metrics exclude user prompts, model responses, internal thoughts and generated SQL.
Agent owners and editors can enable conversation logging for deeper transcript auditing. Google Cloud says privacy checks skip logging when a knowledge source uses row-level security, column-level policy tags or dynamic data masking. For custom agents, the open-source BigQueryAgentAnalyticsPlugin can stream tool-invocation execution trees, time-series telemetry and multimodal payloads into BigQuery.
Cost controls include per-query BigQuery byte scan caps and daily input and output token quotas, with the quotas listed as being in preview. BigQuery logs record input, output, thought, cached and total token counts for each turn, along with total turn latency.
Evaluation moves alongside production monitoring
Google Cloud also describes Prism, an open-source evaluation tool for testing Conversational Analytics agents grounded in BigQuery tables or Looker Explores through the Conversational Analytics API. Prism supports reusable test suites, regular expression assertions and multi-region agent discovery.
The company says Prism can stream completed evaluation runs into day-partitioned BigQuery tables. This places benchmark results alongside production analytics, allowing teams to compare changes to prompts, models or data schemas before they reach users. Google Cloud also points to BigQuery AI functions, the open-source bigquery-agent-analytics Python SDK and a Colab starter notebook for analyzing failures and recurring issues.
For users and platform teams, the announcement brings monitoring, governance, spending controls and pre-deployment testing into the same data environment. The practical effect described by Google Cloud is greater visibility into how agents perform and what they consume, while customer-controlled storage and privacy controls are intended to limit exposure of sensitive conversation data.
Google Cloud: New AI agent observability and governance across BigQuery and Looker


