Accelerating the Speed to Value of Analytics with Databricks AI/BI & Agents

Chris Krysinski18 Jun 2026

Problem

Addepar, a data and AI platform for investment firms, has a 12-person analytics team serving more than 10 business domains. After moving its warehouse from Snowflake to Databricks, it wanted dashboards, a semantic layer and conversational analytics (Genie) in one place. Its semantic models were written in the proprietary language of the previous BI platform, and rewriting them by hand was unappealing. When dashboards compute a metric differently, people argue about whose number is right instead of discussing insights.

Example from the article: AI translated the old semantic models to metric views. Models on tables with metadata came out better on the first pass; models on undocumented data needed much more back-and-forth.

Solution in brief

Keep rich metadata in Unity Catalog and define each metric once as a metric view, so dashboards, Genie and agents read the same logic. Use AI to translate the old models and draft metadata, check every result against the old platform's numbers, and have a human review the metadata. Save what you learn as agent skills.

Questions this case answers
  • How to define each metric once so that dashboards and natural-language agents return the same number, and how to migrate an existing metric layer without changing results
  • How to make a data warehouse understandable to AI agents through governed metadata and reusable procedural instructions that need an owner
  • How to translate a semantic layer and dashboards to Databricks metric views and keep Genie answers accurate?
  • How to make a warehouse understandable to AI agents with metadata and skills?

Tools and methods

  • Metric views — YAML-based semantic layer in Databricks, shared by dashboards, Genie and agents
  • Unity Catalog metadata — descriptions, Markdown links, narrative context
  • Genie, Genie Code (formerly Databricks Assistant), Claude — conversational analytics and AI translation
  • Databricks AI Dev Kit — agent skills and knowledge for Databricks features
  • Agent skills — saved procedures for translation, root cause analysis and dataset discovery
Continue reading — it's free

Google or a one-time email code. No password.