dbt Labs
at Databricks
Data+AI Summit

As AI accelerates the pace of change, demanding fresher data, diverse formats, and support across multiple engines, many teams discover their infrastructure was built for reporting, not real-time AI at scale. Open Data Infrastructure is redefining how organizations design for analytics, operations, and AI. By leveraging Fivetran as an interoperable data foundation, organizations can embrace open standards, separate storage from compute, and keep data portable across clouds and engines, preserving adaptability while scaling:

BI copilots can answer questions about what's on screen: filters, dimensions, chart values. But ask "why is this number wrong?" or "where does this column come from?" and they hit a wall. They can't see past the dashboard. Integral Ad Science built an AI agent that goes deeper. Directly within the BI layer, it connects to your dbt project through MCP and validates findings in Databricks SQL (Unity Catalog) via the Databricks SQL MCP server. The agent extracts context from the report, reads compiled SQL, traces column-level lineage from mart to source, explains the business logic in each transformation, and confirms what the data shows in the warehouse. For analysts, it eliminates hours spent following DAGs and reading model code. For business users, a persona switch delivers clear, natural language answers. Mars Dauer will show how MCP makes this possible across dbt and Databricks and share an end-to-end investigation, from BI report to root cause to Jira ticket, all in one chat.

Building AI apps on Databricks is straightforward. Building AI apps that are trusted, governed, and production-ready is a different challenge entirely. In this talk, we walk through the Databricks Partner Well-Architected Framework with dbt as the data foundation. We'll show how structured, tested, and semantically rich data flows from dbt into Unity Catalog, and from Unity Catalog into any AI application running on Databricks. We'll deploy a live managed app on Databricks that pulls governed assets directly from Unity Catalog with lineage, contracts, and semantic definitions built end-to-end in dbt. The result is an architectural blueprint where your AI apps aren't guessing at the data beneath them; with dbt, they have the required context to be governed, reliable, and right. You'll leave with a Well-Architected reference pattern for AI apps on Databricks, and a clear picture of how dbt and Unity Catalog become the governed data infrastructure every AI use case depends on.