Post by dbt Labs

Your AI app will answer any question with confidence. Whether that answer is correct depends on the data foundation underneath it. Rachael Friedman, solutions architect at dbt Labs, walked through this problem at #DataAISummit and presented a reference architecture for AI apps that are governed and production-ready from the start. Databricks handles the platform and the AI layer. dbt handles the data foundation. That determines whether your app surfaces accurate answers or just confident-sounding ones. What dbt adds to an AI app on Databricks: 🔸 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲: business logic and metric definitions live in version-controlled code, consistent across every app and every query 🔸 𝗧𝗿𝘂𝘀𝘁: full lineage, test status, and freshness signals give every agent interaction an auditable trail 🔸 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: structured metadata means agents and LLMs retrieve the right context faster, with fewer tokens and less compute 🔸 𝗢𝗻𝗲 𝘀𝗲𝗺𝗮𝗻𝘁𝗶𝗰 𝗱𝗲𝗳𝗶𝗻𝗶𝘁𝗶𝗼𝗻: metrics defined once in dbt, so every app queries the same source of truth The reference app they built and demoed: a Databricks Streamlit app powered by the dbt MCP server and governed by Unity Catalog. Teams discover data assets, monitor quality, and query the Semantic Layer, with lineage, ownership, and test status surfaced alongside every result. The AI layer isn't where trust gets built. The data layer is.