A demo video of how AI agents work safely on production data with Bauplan. Using Claude Code with the Bauplan CLAUDE.md and skills, an agent imports an S3 dataset into a Bauplan lakehouse on an isolated feature branch, reviews the schema and preview in the Bauplan UI, and merges to main- no direct writes to production. We then run a downstream pipeline that materializes a metrics table on a new branch with a dry run before merging. Finally, we deliberately break the metrics table to show what happens when a bad write hits prod: one prompt reverts the table to the previous version using lakehouse time travel, and the pipeline can be fixed before the next merge
0:25 — Importing data from S3 into a Bauplan lakehouse
0:33 — Creating a table on a feature branch
1:49 — Merging into the prod branch
2:06 — Verifying data in main branch
3:22 — Merging metrics table into prod
3:28 — Bad pipeline overwrites the metrics table
4:39 — Reverting to the previous version