Microsoft lets AI tools write into Power BI data models
Reviewed by Helen Jones
Microsoft published a Power BI Agentic overview that is not a switch inside Power BI. It is a bundle of agent skills and tools you install so coding agents such as GitHub Copilot can create and edit semantic models and PBIR reports in natural language. Separately, Copilot in web modeling remains in preview in the Power BI service. After a one-time permission prompt it can analyse a model and apply renames, relationships, and DAX measures, and it creates a version-history restore checkpoint first. Copilot cannot be turned on for one workload only. Enabling it on a Fabric capacity for a user or group turns on all Copilot experiences for those people. Identical prompts on an unchanged model may be served from a 24-hour cache.
Until now, most Copilot talk in reporting was about drafting visuals or summaries on top of a model you still owned. The fragile part stayed in Power Query, measures, and refresh. If the AI was wrong, you could ignore the suggestion. The docs now describe a path where natural language can change the live semantic model itself. Microsoft also says unprepared models produce generic or inaccurate Copilot answers. Combined with capacity-wide enablement and a 24-hour prompt cache, a confident-looking edit can land in production, then look unchanged when you re-ask the same question.
Analysis
Treat this as a trap to avoid on reports people already argue about, not a way to skip model work. Before anyone installs Agentic skills or grants web-modeling Copilot, confirm who can write to the workspace and test restore plus a second prompt after a real model change on a copy, so a cached answer cannot hide a bad edit.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "Microsoft lets AI tools write into Power BI data models", Collab365 Spaces. 3 sources referenced.