Data Stack for an AI-Native Company: bauplan at Bauplan

We run our own marketing, sales and product data on Bauplan. In this session Mattia Pavoni opens up that instance and shows the full loop: a business user asks a question in Claude Desktop and if the answer is worth keeping, it becomes a durable pipeline that the rest of the company can use.

He covers the architecture, the repo that governs the agent's behaviour and what happens when Claude decides a question needs a new concept that doesn't exist in the tables yet. From there: a Linear ticket, a synced GitHub issue, a triggered action, and an AI coding agent that opens a data branch, dry runs, and packages a pull request for an engineer to review.

Also in this one: why we don't keep raw data or a gold layer internally, why definitions belong in code and the semantic annotations now written into pipelines and pushed down as metadata on Iceberg tables and columns.