Power BI Development with AI Agents

Power BI Projects as Code

Agents work with files, not with a .pbix. Save your work as a Power BI project and put it under Git, so every change is visible, reviewable and reversible before an agent touches it.

• From PBIX to PBIP: the TMDL and PBIR formats
• Editing outside Power BI Desktop with VS Code
• Git from scratch: commit, diff, branch and revert
• Reading TMDL and PBIR changes
• LAB: Convert a PBIX file to PBIP and put it under Git

Working with a Coding Agent

A chat assistant gives you answers. A coding agent opens your files, makes the change and checks its own work. Learn how to steer it and how much freedom to give it.

• The agentic loop: reason, act, check and repeat
• GitHub Copilot CLI, agent mode in VS Code and Claude Code
• Plan, interactive, auto mode and permissions
• Instruction files: tell the agent about your project
• Choosing a model and keeping token costs under control
• AI Fundamentals: Skills, MCP & Plugins
• LAB: Install an agent and explore a Power BI project read-only

Connecting the Agent to Power BI

Skills tell the agent how to act, tools let it act. Install Microsoft's Power BI plugin and connect the agent to your semantic model with no more access than it needs.

• Skills for Fabric and the Power BI authoring plugin
• The Power BI Authoring MCP server
• Connecting to Power BI Desktop, a PBIP folder and a workspace
• Write and Build permission, XMLA read/write and tenant settings
• What model metadata and query results reach the LLM
• LAB: Connect an agent to a model in Power BI Desktop and in a workspace

From Requirement to Model Plan

A good model starts with the questions it must answer. Let the agent interview you and turn the answers into a plan you review before anything is built.

• Let the agent interview you: business questions, data schema, grain and KPIs
• Star schema proposals
• Working with Modeling Skills
• Reviewing and adjusting the plan before anything is built
• LAB: Write a model spec and build plan for a new subject area

Agentic Semantic Model Development

Describe the change, and the agent builds it. A rename across hundreds of objects takes minutes instead of an afternoon, and the descriptions, display folders and translations nobody likes to write come along in the same run. You stay the architect: the agent proposes, you decide.

• Power Query (M) changes
• Tables, columns, relationships and DAX measures
• Bulk refactoring and renaming across the model
• Descriptions, display folders and translations in bulk
• Deciding on descriptions that define the business
• LAB: Build, refactor and document a model with an agent

Testing and Tuning with an Agent

Agent output is a proposal until it is tested. The agent runs your DAX, compares the results before and after a change, and makes slow measures fast, with the timings to prove it.

• Testing measures with DAX queries
• Comparing results before and after a change
• Server timings with the DAX Studio CLI and the MCP server
• Best Practice Analyzer with the Tabular Editor CLI and the MCP server
• Rewriting slow measures
• LAB: Test and tune a slow model with an agent

Agentic Report Development

From design brief to finished, published pages. The agent writes the PBIR files, reloads Power BI Desktop and studies its own screenshots until the result is right.

• The Power BI Report skill: plan, design, author and manage
• Authoring PBIR pages, visuals, filters and themes
• Replacing legacy visuals and fixing broken ones
• Verifying the result: reload and screenshots through the Power BI Desktop Bridge
• LAB: From design brief to a published report

Agents in Your Team

Microsoft's technical skills know Power BI. They don't know your company. Capture your house rules in a functional skill, review agent changes as a team and settle what the agent may touch before it reaches production.

• Technical vs. functional skills
• Writing a skill for your naming and modeling conventions
• One set of skills for GitHub Copilot and Claude Code
• From one agent to a team of agents: model, DAX and report specialists with one human reviewer
• Pull requests: reviewing agent changes
• LAB: Write a skill for your house rules and review its changes in a pull request

Built-in AI: Copilot for end-users in the Power BI Service

Copilot in Power BI and Fabric IQ mostly serve the people who analyze your data, not the developer who builds the model. See what they add, and how a well-prepared model improves their answers.

• Coding agent vs. built-in Copilot
• Copilot in Power BI Desktop: report pages, summaries and DAX queries
• Copilot in the Power BI service: queries and insights
• Prep data for AI: AI data schema, verified answers and AI instructions
• Fabric IQ and data agents: AI that consumes your model