Design an Agent for BigQuery Analysis

Design an agent that helps users work with BigQuery. Clarify the intended capabilities before choosing an architecture. ### Constraints and Clarifying Questions The reported prompt specifies a BigQuery agent but no dataset, workload, access model, or action policy. For a concrete design, choose a natural-language analysis assistant that discovers authorized schema, proposes read-only queries, executes permitted queries, and explains results. This is an explicit practice scope, not a reported complete product contract. Clarify how metrics are defined, who authorizes execution, and what resource limits apply. ### What a Strong Answer Covers User identity, schema and metric grounding, planning, SQL validation, execution boundaries, cost controls, result interpretation, failure handling, and evaluation. Explain which decisions are enforced outside the language model. ### Follow-up Questions - How would you prevent database content from being treated as instructions to the agent? - What should happen when a query is valid SQL but answers the wrong business question? - How would you recover from an ambiguous execution timeout without blindly starting duplicate jobs?