Momentum One Academy · glossary
Copilot for Microsoft Fabric
Definition
Copilot for Microsoft Fabric (usually documented as Copilot in Fabric) is a family of generative AI assistive experiences embedded in Fabric workloads. It grounds a user’s prompt in permission-scoped context, such as item metadata, schema, or current report state, then sends that request to Azure OpenAI by default. Depending on the workload, it can return guidance, code, SQL or DAX, a visual, or a report draft. It accelerates human work; it is not an authority for data or business decisions. 1 2
Where it sits in Fabric
Copilot is a platform capability, not a Fabric item. It appears in Data Factory, Data Engineering and Data Science, Data Warehouse, Real-Time Intelligence, and Power BI. Administrators control tenant settings, supported paid capacity, and user or workspace access; activity can consume Fabric capacity. 1 3
In Power BI, the semantic model is a key grounding layer: its measures, relationships, names, descriptions, and linguistic metadata give Copilot a business vocabulary. A Fabric data agent is distinct: it is a configurable artifact that queries selected lakehouses, warehouses, semantic models, or KQL databases through read-only, permission-aware conversations. Copilot is preconfigured in a workload; a data agent is scoped and instructed for a use case. 4 5 6
When to use it
Use Copilot for first drafts, code explanations or refactoring, SQL or DAX suggestions, report summaries, and natural-language exploration of a well-prepared model. The user needs access and must be able to validate the result. It is not a sole control for regulated reporting, irreversible changes, or deterministic decisions. 1 2 4
Practical example
For a Module 8 sales solution, an analyst asks Copilot to compare year-to-date net revenue by region using a curated semantic model. Copilot grounds the response in model measures and schema, then returns a visual and explanation. The analyst checks the measure, filters, and totals before sharing. For governed Q&A across a warehouse and KQL database, the team can expose approved sources through a separate Fabric data agent. 4 6
Production caution
Treat every Copilot response as reviewable output. Nondeterministic generation and ambiguous model metadata can produce a plausible but wrong field, filter, query, or explanation. Test representative questions against the production semantic model, keep permissions and governance policies in force, restrict rollout through groups and workspaces, monitor capacity, and confirm regional processing requirements. 2 3 7 8
Module 8 connection
In Module 8, the semantic model supplies business meaning, the data agent provides scoped conversational access, and Fabric governance establishes ownership, access, protection, and audit. Copilot sits above these foundations to speed analysis; it does not replace model design or human review.
Ready to turn these controls into a production-ready pattern? Explore Momentum One Academy’s Fabric Production Workshop.
References
Plan your workshop