BI Insight: Deep Dives into Data, Analytics and Agentic BI

BI Insight: Deep Dives into Data, Analytics and Agentic BI

By Soheil Bakhshi | BI InsightTechnology
Download on the App Store

BI Insight: Deep Dives into Data, Analytics and Agentic BI episodes

  • Microsoft Fabric Apps Business vs Data Apps, Sharing and Licensing

    Microsoft Fabric Apps: Business vs Data Apps – What’s the Difference?

    Microsoft Fabric Apps can be used for quite different scenarios, but the differences between Business Apps and Data Apps go much deeper than the user experience.

    In this video, I compare these two Fabric Apps patterns and look at their architecture, data foundations, sharing mechanisms, semantic model permissions, licensing requirements and Fabric capacity considerations.

    We start with the Business App pattern from my previous video, where I built a transactional application backed by a Fabric-managed SQL database. Then we move to a very different analytical scenario: a Data App built on top of an existing Power BI semantic model.

    In this video, I cover:

    • ​Business Apps vs Data Apps and where each pattern fits
    • ​Fabric-managed SQL databases vs existing Power BI semantic models
    • ​Architecture differences between the two patterns
    • ​Sharing Business Apps and Data Apps
    • ​Semantic model permissions, including Read vs Build considerations
    • ​Execute Queries / DAX API requirements
    • ​Fabric capacity and licensing considerations
    • ​Current preview limitations and gotchas
    • ​When a Business App, Data App, or simply Power BI may be the better fit

    Microsoft Fabric Apps is still in preview, and some behaviours around permissions, licensing and APIs continue to evolve. I discuss some of those nuances in the video as well.

    Listen to the AI-generated podcast:

    Microsoft Fabric Apps Business vs Data Apps, Sharing and Licensing

    Previous Fabric Apps video: I Built a Microsoft Fabric App with GitHub Copilot, Without Knowing How to Code

    📘 My BookExpert Data Modelling with Power BI

    🌐 Follow BI Insight

    Website: BIInsight.com

    YouTube: Subscribe to BI Insight

    Spotify: Follow BI Insight on Spotify

    LinkedIn: Follow me on LinkedIn

    BlueSky: Follow BI Insight on BlueSky

    X: Follow BI Insight on X

    Reddit: Follow me on Reddit

    27 min
  • Agentic AI in Power BI and Fabric, Part 4 From Skills to an Agentic Report Theme Audit Workflow

    Can AI help standardise Power BI report themes across an entire Microsoft Fabric workspace?

    In this episode of the Agentic AI in Power BI and Fabric series, we move beyond working with a single semantic model and explore how Agent Skills can be combined into a much broader agentic workflow.

    Using GitHub Copilot, VS Code, Microsoft report skills, Azure CLI, and Fabric REST APIs, we explore a practical scenario for auditing and standardising Power BI report themes across an entire Microsoft Fabric workspace.

    But the interesting part is not simply applying a theme file.

    We look at how the workflow downloads report definitions, applies an approved theme, validates the changes locally, and deals with real-world complications such as legacy reports and hard-coded visual formatting that prevents proper theme inheritance.

    We also explore how agents, skills, MCP servers, and plugins fit together, why this particular workflow does not require a Fabric MCP server, and how repository-level instructions help keep the agent's behaviour structured and predictable.

    And importantly, publishing changes is not something the agent should simply decide to do by itself. Visual review, audit records, validation, and Human in the Loop approval remain central parts of the workflow.

    The key takeaway from this episode is simple:

    Individual AI skills become much more powerful when they are combined into a structured, repeatable workflow with clear instructions, validation, and human approval.

    📖 Read the full Part 4 blog

    ⬅️ Part 3: Agentic AI in Action with GitHub Copilot and Power BI Modeling MCP

    ➡️ Continue with Part 5: Building a Power BI Theme Audit Agent

    Part 5 takes these building blocks and turns them into a complete, reusable agentic workflow, with exception handling, validation, auditability, and a Human in the Loop approval checkpoint before anything is published back to the Microsoft Fabric workspace.

    This podcast is an AI-generated discussion based on original content written by Soheil Bakhshi. For the hands-on walkthrough, commands, screenshots, and technical caveats, read the original article linked above.

    Follow BI Insight on Spotify to continue the series and explore more content around Power BI, Microsoft Fabric, data analytics, and Agentic AI.

    🎙️ Follow BI Insight

    🌐 BI Insight

    ▶️ YouTube
    💼 LinkedIn
    🦋 BlueSky
    𝕏 X
    🟠 Reddit

    22 min
  • Agentic AI in Power BI and Fabric, Part 5: Building a Power BI Theme Audit Agent

    Welcome to BI Insight.

    In this episode of the Agentic AI in Power BI and Fabric series, all the concepts and building blocks from the previous parts finally come together into a complete Agentic AI implementation.

    The focus is a Power BI Theme Audit Agent designed to inspect reports across Microsoft Fabric workspaces, compare them against an approved theme, identify hard-coded formatting and inconsistencies, apply safe changes locally, create an audit trail, and work through the process towards publishing.

    But the interesting part is not simply whether AI can audit and modify Power BI reports.

    We look at what makes this genuinely agentic rather than just another AI prompt, including reusable instructions, Power BI skills, tools, multi-step reasoning, Human in the Loop approval, and a PreToolUse hook that provides a more deterministic guardrail around publishing.

    And not everything can, or should, be automatically fixed. Legacy report structures, hard-coded formatting overrides, and other exceptions mean the agent needs to recognise when it cannot safely continue and leave the report for human review.

    We also look at the practical side of running an agentic workflow, including the AI credit consumption and approximate cost of running the experiment.

    The key takeaway from this episode is simple:

    A useful Agentic AI workflow is not just about autonomy. It needs tools, context, reusable instructions, validation, deterministic guardrails, and clear points where a human remains in control.

    📖 Read the full Part 5 blog

    ⬅️ Part 4: From Skills to an Agentic Report Theme Audit Workflow

    Part 4 shows how the individual building blocks work together, using multiple Agent Skills to audit and standardise Power BI report themes across a Microsoft Fabric workspace. Part 5 takes those building blocks and turns them into the complete, reusable Theme Audit Agent.

    This podcast is an AI-generated discussion based on original content written by Soheil Bakhshi. For the hands-on walkthrough, implementation details, screenshots, prompts, and technical caveats, read the original article linked above.

    Follow BI Insight on Spotify to continue exploring practical Agentic AI, Power BI, Microsoft Fabric, data analytics, and governance.

    🎙️ Follow BI Insight

    🌐 BI Insight

    ▶️ YouTube
    💼 LinkedIn
    🦋 BlueSky
    𝕏 X
    🟠 Reddit

    14 min
  • I Built a Microsoft Fabric App with GitHub Copilot, Without Knowing How to Code

    This episode is the AI generated audio version of the article and video of the same name.

    Can someone who does not write code build a working Microsoft Fabric App? I spent about four and a half hours finding out, with GitHub Copilot doing the typing and me doing the thinking. What came out the other end is an AdventureWorks style operational sales app: customers, products, inventory, sales orders, an approval workflow and a reporting page, sitting on a Fabric SQL Database and deployed as a Fabric App.

    In this episode

    • The Ask, Plan, Human in the Loop, Agent workflow, and why the order matters more than the prompts
    • Grounding Copilot in the official Fabric docs with #fetch before it writes a single line
    • Using #create-instructions to lay down guardrails that survive across sessions
    • Scaffolding with the Rayfin SDK and CLI, and the moment rayfin up quietly deployed live resources
    • Building the schema, the CRUD screens and the approval workflow phase by phase, with approval gates
    • Testing with real users, then deploying and sharing the app (and why nobody gets an email)
    • Where Fabric Apps sits next to Power BI and Power Apps, and why it replaces neither

    The honest bit: AI generated code is still code. It has bugs, it needs testing, and domain knowledge is what tells you whether the data model and the business logic are actually right. A fast first implementation is not the same as a finished solution.

    Read the article
    https://biinsight.com/building-a-microsoft-fabric-app-with-github-copilot-without-knowing-how-to-code/

    Watch the video
    https://youtu.be/VDsU-z27wkY

    More at biinsight.com

    18 min
  • I Built a Microsoft Fabric App with GitHub Copilot, Without Knowing How to Code

    Can you build a real Microsoft Fabric App with GitHub Copilot, even if you are not a developer?

    In this video, I build an AdventureWorks transactional application from scratch using Microsoft Fabric Apps, the Rayfin SDK, Visual Studio Code, and GitHub Copilot.

    Rather than simply asking Copilot to generate an app, I use a structured AI-assisted workflow: define the context, ground Copilot with Microsoft documentation, create repository instructions and a PRD, build an implementation plan, then let the agent implement it in controlled phases with human approval checkpoints.

    The result is a working Fabric App with customers, products, inventory, sales orders, SQL database, authentication, CRUD operations and reporting, deployed to Microsoft Fabric.

    I am not a software developer. This is vibe coding, but with structure, grounding and guardrails around the AI rather than simply generating an app and hoping for the best.

    🚀 In this video, you will learn
    • ✅ What Microsoft Fabric Apps are and where they fit
    • ✅ How to enable Fabric Apps and set up prerequisites
    • ✅ How to install and use the Rayfin CLI
    • ✅ How to use GitHub Copilot in VS Code for vibe coding
    • ✅ How Ask, Plan and Agent modes fit into the workflow
    • ✅ How to create instructions, a PRD and implementation plan
    • ✅ How Copilot can build and test a Fabric App
    • ✅ How to deploy and share a Fabric App
    • ✅ Where Fabric Apps fit alongside Power BI and Power Apps
    🌐 Complementary Blog

    Building a Microsoft Fabric App with GitHub Copilot Without Knowing How to Code

    📚 Agentic AI in Power BI & Microsoft Fabric

    Part 1 – Concepts, Terminology, and How to Think About It
    Read Part 1

    Part 2 – VS Code, GitHub Copilot, and Safe MCP Setup
    Read Part 2

    Part 3 – GitHub Copilot and Power BI Modeling MCP
    Read Part 3

    Part 4 – Agentic Report Theme Audit Workflow
    Read Part 4

    📖 Microsoft Fabric Apps Documentation

    Fabric Apps documentation
    What is Fabric Apps?
    Deploy a Fabric App

    📕 My Book

    Expert Data Modeling with Power BI, 2nd Edition

    🔗 Connect With Me

    Blog: biinsight.com
    YouTube: @BI.Insight
    LinkedIn: linkedin.com/in/bakhshi
    BlueSky: biinsight.com
    X: @_SoheilBakhshi

    💬 What Do You Think?

    Where do you see Fabric Apps fitting? Could they replace Power BI or Power Apps for some use cases, or are they a different workload?

    If you found this useful, please like, subscribe, and share your thoughts in the comments.

    28 min

About BI Insight: Deep Dives into Data, Analytics and Agentic BI

From the publisher's feed

BI Insight is a technical show about data, analytics and agentic BI, hosted by Soheil Bakhshi, Microsoft Data Platform MVP.