Google Cloud introduces Data Agent Kit, letting coding agents build BigQuery, Spark and Airflow data pipelines

Google DeepMind

Google Cloud introduced Data Agent Kit, a plugin for coding agents and an extension for VS Code-based IDEs that gives agents tools and skill files for its data services. Its worked example has an agent generate Spark notebooks, a dbt project and an Airflow pipeline for a fraud-review workflow. The post tells readers to review generated code before running it.

• Two parts: MCP tools that connect an agent to data services, and Skills, which are Markdown files with instructions and best practices for a data stack.
• The example ingests JSON transaction logs into BigQuery, builds a dbt project that deduplicates records, and trains a RandomForest classifier on Spark Serverless.
• Transactions scored at 0.50 or higher are written to a Spanner review table; that threshold is the example's choice, not a product default.
• The model only flags transactions for review; the post says they still need to determine whether fraud occurred.
• The post gives no pricing. Besides VS Code forks such as Antigravity and Cursor, it lists plugins for Claude Code and Codex.