Data science agent that plans and runs a notebook
Upload a dataset and describe a goal; an agent plans the analysis and writes a working notebook you can inspect and rerun.
The use case
Students, researchers and analysts often know the question but not every step of the pandas or scikit-learn code. Google's Data Science Agent in Colab takes a dataset and a goal, such as 'find what predicts churn', writes a plan, and generates a full notebook with cleaning, analysis and charts that runs in Colab. Because the output is a regular notebook, every step can be read and changed. The person still checks data cleaning choices, looks for leakage in any model, and does not treat results as final without review.
Example workflow
- Upload the dataset to Colab
- Describe the goal in plain language
- Review the plan the agent writes
- Run and read the generated notebook
- Fix assumptions and rerun
Tools
- Google Colab
- Gemini
Risks and guardrails
- Cleaning choices like dropping rows can bias results
- Model results can suffer from leakage
- Do not upload restricted data to personal accounts
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Google Developers Blog: Data Science Agent in Colab: developers.googleblog.com/en/data-science-agent-in-colab-with-gemini
- Gemini (official page): gemini.google.com (checked 2026-11-09)