Microsoft used its Build 2026 developer conference in San Francisco to announce a fundamental shift in how Power BI works. The headline: Agent Skills for Power BI, now in preview, gives AI agents the ability to handle the full analytics development cycle — from raw data to semantic model to published report — based on nothing more than a plain-language description or a screenshot of what you want.

For anyone who has spent hours building DAX measures, managing table relationships, or iterating on visuals to match a stakeholder’s expectations, this is a material change.

What Agent Skills for Power BI Actually Does

The announcement covers two capabilities released together at Build 2026.

Agent Skills for Power BI (Preview) provides what Microsoft calls “true end-to-end agentic development.” You describe what you need — in plain language, or by pointing the agent to a screenshot of an existing report — and the agent handles construction. It builds the semantic model from your Fabric data, writes the DAX, generates report pages, and iterates on visuals while aligning to design best practices. The goal is a complete, usable output with minimal manual work at each stage of the pipeline.

This is meaningfully different from Copilot features already embedded in Power BI, which assist a human doing the work. Agent Skills is designed to complete the work autonomously within the constraints of a development workflow.

Fabric Apps for Semantic Models extends the concept beyond reporting into custom applications. Using a coding agent and a well-built semantic model, developers can now generate complete Fabric-native web apps — financial planning tools, inventory dashboards, pricing optimisation interfaces — using natural language specifications. The semantic model provides the trusted business logic; the agent handles the front-end build.

The Semantic Model Is Now the AI Interface

What makes both capabilities work is the semantic model itself. Microsoft made this explicit in the Build announcements: a well-curated semantic model — with clean measures, defined relationships, proper hierarchies, and business terminology attached to the right columns — is now the layer that allows AI agents to reason correctly over your data.

An agent building a report on top of a poorly structured semantic model will produce unreliable output. An agent working from a properly governed model can produce trustworthy analytics because it inherits the business logic that data professionals have already encoded.

This inverts a concern that has followed the rise of AI analytics tools: that AI would make skilled data modelling less relevant. The opposite appears to be true. The better the model, the more the agent can do with it.

Fabric Skills for Developer Tooling

Also announced at Build: Fabric Skills for GitHub Copilot, Claude, and CLI — a set of skills built by Microsoft and open for community contribution that allow AI coding agents to interact with Fabric workloads directly. Developers using GitHub Copilot or Claude can now query Fabric data, push model changes, and manage workspace assets without leaving their development environment.

This mirrors the Power BI Modeling MCP that Microsoft released in November 2025, which first made Power BI semantic models accessible to MCP-compatible agents. Build 2026 builds on that foundation by expanding the capability surface and packaging it for the most common developer tools.

What This Means for Business

If you manage a Power BI environment: Agent Skills does not replace your semantic models — it depends on them. Teams that have invested in proper data modelling, governance, and consistent business definitions are now sitting on assets that AI agents can use immediately. Teams that have not made that investment will find the agents produce inconsistent output until the model is cleaned up.

If you are a data professional: The workflow is changing, not disappearing. Describing the requirement and reviewing agent output is a different skill from building everything from scratch — but it still requires understanding what a good semantic model looks like and how to verify that generated DAX is actually correct. The demand for people who can do this well is likely to increase, not decrease.

If you are commissioning BI work: Timelines for report development should shorten significantly as these tools move from preview to general availability. The cost-per-report equation changes when a well-specified agent can generate a draft in minutes. The bottleneck shifts to defining the requirement clearly and reviewing output critically.

If you are evaluating AI readiness across your business: The Power BI announcements at Build 2026 are a practical signal of where enterprise AI is heading. AI agents are moving into the production layer of every major business tool, not just as assistants but as autonomous builders. The organisations that have already invested in clean data foundations are the ones positioned to take advantage of that shift immediately.

The Databricks Data + AI Summit runs June 15-18 in San Francisco, where further announcements across the agentic analytics space are expected. For data and AI professionals, the pace of change in the tooling layer this month is unusually high.

---

Enterprise DNA has trained more than 220,000 data professionals on Power BI, DAX, and data modelling. Explore EDNA Learn to build the semantic modelling skills that make agentic analytics work — or talk to us about Omni if you want to deploy AI agents across your business operations.