Fail Before You Run: Typed Data Contracts for AI-Written Pipelines

Your coding agent writes a pipeline. It runs for an hour, then fails because a column it assumed was an integer is now a float. Or it succeeds and a downstream table breaks later.

In this webinar, Ciro Greco and Giacomo Piccinini show how Bauplan 0.3 moves those checks into the code. Every model declares typed table schemas for what it reads and returns: column types, nullability and documentation, versioned with the pipeline. A mismatch stops the run at the contract check, before the DAG runs and before anything reaches main.

In the live demo, an upstream system changes loyalty points from int64 to float64 in month three. The import fails at the contract check on an ingestion branch, main stays untouched, and Claude fixes the code on a separate branch and opens a PR for review.

What we cover:

• Why developer tools built for humans need rethinking when the user is an agent
• Typed table schemas in Python and SQL, and why they give you data contracts for free
• Token efficiency and failing fast: checking before you pay for compute
• Contracts vs expectation tests, and why you keep both
• Documentation that lives in the code and in Iceberg metadata

Blog post: https://bauplanlabs.com/post/devex-for-developers-who-dont-write-code
‍Slide deck: https://drive.google.com/file/d/1_i9rKsC8QKX6UK7B14co_Hu46MEdp-jZ/view