Google previews QueryData as an agentic data layer for Gemini/BigQuery agents

Google Cloud’s new QueryData preview is a database tool for AI agents, not a new BigQuery layer. Google launched it on April 10 for AlloyDB, Cloud SQL, and Spanner. (cloud.google.com) QueryData takes a natural-language prompt and returns generated SQL, an explanation of intent, optional query results, a natural-language answer, and a follow-up question if the request is ambiguous. Google exposes it through the Conversational Analytics API. (docs.cloud.google.com 1) (docs.cloud.google.com 2) The product pitch is accuracy. Google said QueryData uses “context sets” — structured business logic and examples supplied by developers — to translate plain-English questions into database queries with “near-100% accuracy.” (cloud.google.com) (docs.cloud.google.com) That matters because large language models are usually good at writing SQL syntax but weaker at understanding what a company’s tables, abbreviations, and business rules actually mean. Google framed the main risks as wrong answers, weak access controls, and heavy setup work. (cloud.google.com) BigQuery is part of the same broader agent story, but through different features. Google’s BigQuery docs now describe preview “data agents” that let users chat with selected tables, views, and user-defined functions using natural language. (docs.cloud.google.com 1) (docs.cloud.google.com 2) Those BigQuery agents rely on metadata, instructions, and “verified queries,” formerly called golden queries, to steer answers. Gemini in BigQuery also already offers data insights, data canvas for finding and joining tables, SQL and Python assistance, and AI-guided data preparation. (docs.cloud.google.com 1) (docs.cloud.google.com 2) Google tied QueryData to its database announcements at Google Cloud Next ’26, where it described “Tools for Data Agents” as modular building blocks for custom agents. In that lineup, QueryData is the high-accuracy text-to-SQL function for operational databases. (cloud.google.com) The caveat is that both QueryData and BigQuery data agents are still preview products. Google’s documentation says pre-General Availability features are offered “as is,” may have limited support, and should be validated before production use. (docs.cloud.google.com) (docs.cloud.google.com) So the cleaner read is narrower than the original social-media framing: QueryData is Google’s new agentic query engine for AlloyDB, Cloud SQL, and Spanner, while BigQuery has its own parallel agent and Gemini features for analytics work. (cloud.google.com) (docs.cloud.google.com)