“Bringing you the Data Agent Kit” describes an end-to-end fraud-investigation workflow built with the Data Agent Kit inside the Antigravity IDE. The kit works with Google Cloud data through a plugin for coding agents or an extension for Visual Studio Code-based IDEs. It contributes MCP tools that connect agents to data services and Markdown skills that give agents instructions and best practices for the data stack. In the IDE, the kit also provides tools for Google’s Agentic Data Cloud, from querying databases to deploying an Airflow pipeline without leaving the editor.

The example starts with a compliance team noticing payments are taking longer to clear and suspecting fraud. The data scientist needs a shortlist of transactions to investigate. Raw transaction logs sit in Google Cloud Storage. The first prompt asks the agent to create a notebook that loads them into BigQuery using Spark. The agent generates a Jupyter notebook that writes the JSON logs to a BigQuery table named raw_transactions. You can open the notebook, review the code, and run it with a preconfigured Spark Serverless runtime. After it runs, the Data Agent Kit extension lets you browse the table and inspect its metadata directly from the IDE.

Those records are in BigQuery but not ready for training. The next prompt asks for cleanup logic in a dbt project so it can be rerun on every new batch. The agent generates SQL models for cleaning and enriching the data, plus a uniqueness test for transactions. Running dbt build in the integrated terminal builds staging and enriched tables, creates a separate table for records with missing transaction IDs, and runs the uniqueness test. The is_fraud column is kept so past labeled records can train a model and unlabeled new records can be scored later.

To build the review queue, the agent creates a training notebook using a Random Forest classifier on Spark Serverless. After running it, you check the evaluation output to see how well the model separates fraudulent transactions from legitimate ones. The extension lets you verify that trained model artifacts were saved to Google Cloud Storage. Next, an inference notebook scores the unlabeled transactions in BigQuery and sends those with an estimated fraud probability of 50% or higher to a Spanner review queue. You can query the Spanner table in the IDE to inspect the flagged records. The compliance team can then decide where to investigate, though they still need to determine whether fraud actually occurred.

Finally, the post automates the daily run. The agent generates a YAML pipeline definition and deployment configuration. The definition sets the schedule and dependencies between steps; the configuration specifies the Managed Service for Apache Airflow environment. Data Agent Kit turns the definition into an Airflow DAG shown as an interactive graph in the IDE under “Orchestration Pipelines.” Clicking a task links back to notebook code. With the Airflow environment configured, clicking “Run pipeline” packages the notebooks and dbt project, deploys the DAG, and triggers a run. Execution history and per-task status, duration, and logs are visible, and Airflow runs the pipeline every day on new batches.

The post closes by pointing readers to a codelab for the full setup, plus documentation and the open-source repository on GitHub for the skills and tools.