In Short: The Lakehouse Grew Its Own BI, Then Made It Conversational
Databricks spent years as the platform BI tools connected to; AI/BI is its bid to be the platform BI happens on. Two pieces: AI/BI Dashboards, a modern dashboarding experience over SQL warehouses with AI-assisted authoring, and Genie, conversational analytics - business users asking questions in natural language against governed lakehouse data, with the generated SQL shown rather than hidden.
For anyone tracking the agentic analytics space, Genie is Databricks' entry in the same race Microsoft runs with Copilot and Fabric IQ: grounding conversational answers in governed enterprise data.
AI/BI Dashboards, Briefly
The dashboards are what you would expect from a 2020s rebuild: SQL-warehouse-backed, fast over Delta tables, shareable to non-Databricks users, with AI assistance drafting charts and summaries. For estates already serving analysts in Databricks they remove a tool hop; estates committed to Power BI treat them as complementary operational views rather than a replacement, and the Power BI patterns remain the enterprise reporting spine.
The more strategically interesting half is Genie.
How Genie Actually Works
A Genie space is a curated conversational environment, and the curation is the product.
- Scope: you choose the Unity Catalog tables the space may see - a bounded domain, not the whole lake
- Instructions: plain-language rules teach the space your definitions: which revenue is authoritative, how fiscal periods work, what to refuse
- Example queries: worked question-to-SQL pairs that anchor how your organisation's questions map to your schema
- Transparency: every answer shows the SQL it ran, so a sceptical analyst can verify rather than trust
Governance rides along by construction: queries execute under Unity Catalog permissions, so answers are filtered to the asker's access, and lineage and audit capture the traffic.
If that recipe sounds familiar, it should: it is the same curation discipline as Fabric data agents - scoped sources, instructions, examples, an owner. The industry has converged on the shape of trustworthy conversational analytics; the platforms differ on where the grounding lives.
Genie or Copilot? The Real Decision
The honest comparison is not feature-by-feature; it is about where your semantic truth and your audience live.
- Definitions in Power BI semantic models, users in Teams and Power BI: Copilot grounded through Fabric IQ is the natural surface, and model preparation is the work
- Definitions in Unity Catalog, users in the Databricks estate: Genie spaces are the natural surface, and space curation is the work
- Both, honestly: common in dual-platform estates - engineering asks Genie, the business asks Copilot, and the same Delta tables back both
The failure mode is identical on both sides: enabling the conversational layer over an uncurated estate and letting fluent wrong answers burn trust. The readiness work is platform-neutral, which is the most reassuring thing we can tell any client weighing this space.
Sources and Further Reading
Frequently asked
The umbrella for Databricks' business-intelligence experiences: AI/BI Dashboards (a modern dashboarding tool with AI-assisted authoring over SQL warehouses) and Genie (conversational, natural-language analytics). Both run on the lakehouse directly, governed by Unity Catalog.
A curated conversational environment: you select the tables, add instructions and example queries that teach it your business definitions, and business users then ask questions in natural language. Genie generates and runs SQL against the governed data and shows its work.
Same category, different grounding: Copilot resolves questions against Power BI semantic models; Genie resolves them against Unity Catalog tables with curated instructions. Which answers better in your estate depends on where your definitions and governance actually live.
Yes - queries run under Unity Catalog governance, so users get answers only from data they can access, and the generated SQL is visible for verification. Curation quality, not permissions, is usually the limiting factor on answer quality.
In practice estates pick a primary conversational surface per audience: Genie for teams living in Databricks, Copilot for those in Teams and Power BI. Because both platforms share Delta data, the same governed tables can back both - the differentiator is where each audience already works.
