Google's Data Agent Kit links coding agents to BigQuery

The announcement sits alongside Google's Knowledge Catalog, previously known as Dataplex, which is described as a "context engine" for metadata, data lineage, and data quality across BigQuery, AlloyDB, Cloud Storage, and other sources. Paired with Data Agent Kit, the claim is that a coding agent can look up internal data structures on its own rather than being fed schemas and error logs by hand.

The stated problem is that ordinary AI agents lack context about which tables exist and how they are partitioned. Through the kit, an agent can search Knowledge Catalog so a developer can ask in plain language — for example, predicting next month's demand for best-selling products — and have the agent identify trusted sales and inventory tables. Google also says the kit bundles open-source skills written by its data engineers, so an agent checks partition keys before running a BigQuery query to avoid an accidental full-table scan and its high charges.

The broader framing is a "SOAP" (Semantic and Ontology Agent Platform) architecture combining three products: a borderless lakehouse for zero-copy access to data in other clouds such as AWS S3, Knowledge Catalog for context, and Data Agent Kit for automated development. This is a response to three challenges — fragmented and varied data, rising token consumption from complex AI tasks, and governance — and GA for Data Agent Kit is said to have begun in October 2026.