AI didn't change our dbt project, just who cared about it.

The day after Starburst announced our in-product AI analytics agent internally, our executive team challenged us to point it at our own data, as soon as possible. Our agent hit production and 45 days later, we shipped our first internal phase to the entire leadership team. Our secret? Our bronze and silver layers were already curated in a dbt project we used daily, so we spent our time tuning with subject matter experts and connecting our dbt descriptions downstream to the agent.

I'll cover the guardrails we imposed once executives cared what lived in our dbt repo (including the one that saved us: no build starts until the business unit commits), how dogfooding shipped changes for all customers, like extending persist_docs in dbt-trino, and how we're scaling as more teams onboard. You'll leave with the rules we'd keep, the mistakes we'd skip, and how we're making this easier for others today. It turns out the best sponsor our dbt repo ever had turned out to be an AI project.