Skip to content

THR926 Theater Thu 19 Nov 16:20 PST

Power BI Development in the Agentic Era

Sujata Narayana

Summary

From the recording

Sujata Narayana demonstrated agentic Power BI development using the GitHub Copilot app. The toolkit comprised a Power BI authoring plugin, a Fabric apps plugin and the Work IQ MCP, plus Power BI Desktop for report validation. The authoring plugin is not default-enabled yet; Microsoft says it will become so. She advocated spec-driven development, choosing models by task complexity, and human validation. A published semantic model can also power a custom Fabric app, which took roughly 20 minutes to build in the demo. No GA or preview dates were given for the plugins or Fabric apps.

Session description

Microsoft's catalogue description

Power BI development has entered a new phase that is shaped by agentic experiences. Join the product team to see how these capabilities are changing the way data solutions are built and improved, from semantic models to reports and even custom web apps. Learn how to best leverage Al agents, like GitHub Copilot, to build more powerful data experiences with less manual effort.

Demo timestamps

Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.

  1. Skip to 00:07:19

    Setting up GitHub Copilot app for Power BI authoring

    In the GitHub Copilot app, Work IQ MCP is already present under Customise. The presenter searches for the Raven Power BI authoring plugin and uses Power BI Desktop so the agent can screenshot and validate reports.

  2. Skip to 00:11:51

    Agent produces a markdown requirements file

    The agent returns a markdown requirements file covering purpose, KPIs, dimensions and critical definitions.

  3. Skip to 00:16:22

    Resulting report and semantic model

    The generated report matches the design standards and logo. The model view shows the semantic model the agent built.

  4. Skip to 00:19:47

    Handing the spec to a coding agent to build a Fabric app

    The same spec and design standards are given to the coding agent to create a Fabric app. The speaker says this took about 20 minutes off-camera.

  5. Skip to 00:20:47

    Polished Fabric app shown

    The more polished Fabric app produced by the coding agent is shown.

  6. Skip to 00:21:21

    Vibe-coding write-back and annotations

    Write-back and annotation features are added to the app by vibe-coding.

  7. Skip to 00:23:29

    Connecting an OpenAI endpoint without exposing API keys

    Shows how to point the agent at an OpenAI endpoint without putting API keys in chat or markdown. A security scan is recommended, and GitHub Copilot's built-in scanner is mentioned.

  8. Skip to 00:24:50

    Finished pane generates ideas with an LLM

    The finished pane generates ideas from an LLM, using data insights and a collaborator comment.

Watch on-demand (opens in a new tab)

← All sessions