For years, the tedious part of working with Power BI wasn't building reports. It was semantic modeling: renaming tables, setting up relationships, debugging DAX measures that should have been straightforward but ended up eating an entire afternoon. Microsoft decided to put Copilot right there, in the middle of that pain.
The June 2026 update brought two features that change how analysts and developers actually work with Power BI: Copilot in Web Modeling and Agent Skills. Both are in Public Preview, but the direction is already clear.
What Copilot does inside the semantic model
The concept is straightforward: you open a semantic model in the Power BI service, switch to editing mode, and call Copilot from the ribbon. From there, you type what you need in plain language.
"Create a profit margin measure." "Rename the tables to follow fact and dimension naming conventions." "Check if there are any relationships that might be broken."
Copilot analyzes the schema, flags inconsistencies, and applies the changes directly. Before doing anything in your first session, it asks for permission and creates an automatic restore point so you can roll back if something goes sideways.
That's not a small thing. Anyone who has spent 30 minutes figuring out why a measure returns BLANK instead of zero will appreciate the value here.
- Model analysis with best-practice recommendations
- Table and column renaming
- Relationship creation and modification
- DAX measure generation from text descriptions (totals, growth metrics, aggregations)
The Copilot input limit jumped from 500 to 10,000 characters, so you can provide real business context instead of vague one-liners.
Agent Skills: AI that builds from scratch
Agent Skills sit a level above Copilot. Announced at Microsoft Build 2026, they aim at something more ambitious: an AI agent that manages the full analytics development cycle, from raw data to published report, driven by a natural language description or even a screenshot.
Here's how it works: you describe what you need. The agent creates the semantic model (star schema, DAX measures, AI-ready configurations), builds the report, formats it, and publishes it. Without you clicking on a single table.
Microsoft is deliberate about distinguishing the two: Copilot in Web Modeling assists the human editing an existing model. Agent Skills are for autonomous development, where the agent takes the lead on the entire process.
For teams with high volumes of repetitive operational reports, combining both could mean a significant drop in delivery time.
DAX User-Defined Functions: now generally available
While Copilot and Agent Skills grab the spotlight, the DAX User-Defined Functions (UDFs) reaching general availability deserves attention.
UDFs let you define a complex calculation once and reuse it across measures, calculated columns, and visual calculations. This addresses a long-standing pain in Power BI: duplicated calculation logic scattered across the model, making maintenance harder and inconsistency more likely.
If you have a contribution margin or budget-vs-actual formula copied across ten different measures, you can now centralize it in one function and call it wherever needed.
The semantic model as the center of everything
What Microsoft is building, piece by piece, is an architecture where the semantic model is the trusted business intelligence layer that feeds both human-made reports and AI agents.
The June update reinforces this with the "Prep data for AI" experience, which evaluates semantic model quality specifically so AI-generated answers are more accurate. There is also the Outbound Access Control (OAP) in Fabric, which lets workspace administrators restrict which data sources models and reports can connect to.
The logic is clear: the better the model, the more reliable everything the AI produces from it.
What this means in practice
For Power BI analysts and developers, the landscape is shifting. Not in the sense of "AI will replace you," but in the sense that repetitive and mechanical modeling tasks are being absorbed by AI.
That means the professional edge will increasingly come from business understanding: knowing which question to ask the model, which measure has real semantic meaning, which report actually helps a decision get made. The technical side of "writing correct DAX" is becoming a commodity.
Those who understand the underlying concepts, like star schemas, granularity, and DAX filter context, will use Copilot as a productivity multiplier. Those who always copied formulas from Stack Overflow without understanding them will keep copying, just in a more sophisticated way.
The features are in Preview. To enable them, you need a Microsoft 365 Copilot license and Fabric capacity. Official documentation is at learn.microsoft.com/en-us/power-bi/transform-model/copilot-web-modeling.