The article details the integration of Looker‘s governed semantic layer with Gemini Enterprise, enabling trusted self-service business intelligence through natural language queries.

The problem addressed is that large language models often hallucinate on raw enterprise databases, while NL2SQL models can produce inconsistent metrics.

Looker‘s semantic layer eliminates this guesswork by serving codified, version-controlled business logic to generate deterministic, precise SQL.

The integration uses the Agent-to-Agent (A2A) protocol, ensuring that user requests are routed to a Looker agent, which returns governed metrics without data ingestion or storage.

Key features include rich visual interactivity with charts, interoperability with first- and third-party agents for multi-agent orchestration, and robust governance via OAuth-based user authentication that enforces existing Looker permissions, row-level filters, and column-level masking.

The architecture is a zero-risk pass-through, meaning Gemini Enterprise does not ingest or store underlying database records.

This allows organizations to provide a unified AI interface for both structured and unstructured data, reducing friction to adoption and promoting a data-driven culture.