Databricks Brings Genie Into Microsoft Copilot for Governed Business Data
Databricks is wiring its Genie assistant into Microsoft Copilot Cowork, Teams, and Excel so business users can query governed enterprise data without leaving the Microsoft 365 stack.
Takeaways
- Databricks launched an Azure Databricks Genie plugin for Microsoft Copilot Cowork and Copilot Chat.
- The Databricks Genie app for Teams is now generally available with DM and @mention support.
- Copilot delegates data questions to Genie One, which answers using Genie Ontology and Unity Catalog governance.
- Excel Add-in adds live pivots on metric views, scheduled refreshes, and write-back to Databricks tables.
- Every query runs under the signed-in user's permissions; the Teams routing layer stores no questions or answers.
- Work IQ connectivity over MCP is coming, bringing Microsoft 365 context into Genie answers.
Databricks has expanded Genie across Microsoft 365 with a new Azure Databricks Genie plugin for Microsoft Copilot Cowork and Copilot Chat, general availability of its Teams app, and broader capabilities in its Excel add-in. Developers can also connect Genie Agents to custom Copilot Studio assistants through the Azure Databricks connector for Power Platform.
Enterprise assistants can read email, files, and calendars, yet often lack a company’s approved definitions for metrics such as revenue or net retention. Databricks addresses that gap by routing data questions through Genie, where Unity Catalog permissions and shared business definitions govern the response. The approach reduces reliance on raw column names and model inference.
Copilot hands questions to Genie
- Route the request: Copilot sends governed-data questions through the plugin to Genie One, Databricks’ assistant for querying data and business context across an organization.
- Resolve the language: Genie Ontology maps the user’s wording to approved metrics, dimensions, and related business concepts.
- Check access: Unity Catalog applies the requester’s permissions before Databricks queries the underlying data.
- Run the query: Databricks executes the request against authorized sources and generates an answer from the results.
- Return the response: The answer appears in the Microsoft interface where the question originated.
Genie Ontology combines explicit definitions from analytics teams with context derived from the use of dashboards, queries, and documents. Metric views store shared calculations such as annual recurring revenue, while certified dimensions standardize categories such as region or customer tier. This semantic layer gives Genie a consistent interpretation of business language before it queries data.
One back end, four Microsoft surfaces
| Surface | Availability and setup | Capabilities |
|---|---|---|
| Copilot Cowork and Chat | Users enable the plugin and authenticate with Azure Databricks credentials. The setup requires no OAuth app registration or manifest edits. | The Genie plugin routes business-data questions to Genie One. |
| Microsoft Teams | The Teams app is generally available. | Users can message Genie directly or mention @Databricks Genie in a channel or group chat. Genie replies in the thread. |
| Excel | The Databricks Excel add-in connects workbooks to governed data. | Analysts can ask questions in natural language, receive cited answers, create live pivots from Unity Catalog metric views, schedule refreshes, and write forecast inputs to Databricks tables. |
| Copilot Studio | Builders connect through the Azure Databricks connector for Power Platform. | Custom assistants can invoke Genie Agents for governed queries and workflows. |
Databricks wants the semantic layer
Azure Databricks has linked Microsoft’s cloud and Databricks’ data platform for years. Generative AI adds a new requirement: assistants need an authoritative source for metric definitions, permissions, and business terminology. Microsoft Fabric offers its own semantic models, while Databricks is positioning Genie Ontology and Unity Catalog as the corresponding layer for organizations that already manage data in Databricks.
Placing Genie inside Copilot allows Databricks-defined metrics to reach Microsoft 365 users without a separate implementation in Fabric. For organizations comparing the platforms, the integration affects architecture as well as user experience because the selected semantic layer determines where teams maintain definitions, access policies, lineage, and audit controls.
Agents can act after approval
Genie in Teams can invoke tools through Model Context Protocol connectors. MCP is an open protocol that lets AI systems call external tools and retrieve data through defined interfaces. With compatible connectors, Genie can create or update a document, comment on a ticket, or change a record. Databricks says Genie requests the user’s approval before committing a change.
Permissions follow the requester
Questions submitted through Copilot and Teams may contain sensitive business information, so the design preserves Databricks identity, networking, and policy controls across the Microsoft interfaces.
- User-level execution: Each request runs as the person who submitted it. Unity Catalog and source-system permissions determine which data Genie can access.
- Workspace processing: Databricks processes questions inside the workspace. The relay service between Teams and Databricks does not store questions or answers.
- Private networking: Private Link workspaces are supported through a context-based ingress policy configured by a Databricks account administrator.
- Administrative controls: Unity Gateway provides request traces, MCP policy enforcement, and usage budgets for AI workloads.
The roadmap pulls in Work IQ
Databricks says a forthcoming MCP connection to Microsoft Work IQ will bring Microsoft 365 context into Genie, including email, meetings, documents, Teams messages, directory information, and organizational relationships. A single request could then combine governed warehouse metrics with recent customer correspondence or meeting context. The company has not specified a release date.
What teams still need to test
Production quality will depend on the definitions, permissions, and source data that each organization maintains. Stale metric views or incomplete business terminology will produce weak answers even when the request path is correctly governed.
- Identity mapping: Confirm that Microsoft identities resolve to the intended Databricks principals and test row-level and column-level restrictions.
- Semantic accuracy: Compare Genie’s answers with approved finance and operations reports before broad deployment.
- Agent boundaries: Limit MCP tools to required actions and verify approval behavior for every write operation.
- Network policy: Validate Private Link ingress rules and workspace connectivity from each Microsoft surface.
- Operations: Review traces, usage budgets, retention settings, and incident procedures for assistant requests.
Existing Azure Databricks customers can reuse their Unity Catalog policies and Genie definitions across several Microsoft interfaces, reducing duplicate integration and semantic-model work. The same architecture ties Microsoft 365 workflows more closely to Databricks’ ontology, catalog, and agent controls, making those components part of the long-term platform decision.