AI Daily | Microsoft Fabric
The Announcement
At Build 2026, Microsoft quietly shipped one of the more consequential pieces of the Fabric roadmap: Rayfin, an open-source SDK and CLI for building real, code-first web applications natively inside Microsoft Fabric. Not dashboards wrapped in an iframe. Not a Power Apps canvas bolted onto a dataset. A genuine TypeScript web app, with its own backend, database, and authentication — deployed as a first-class Fabric item.
In a recent walkthrough, Microsoft’s Patrick LeBlanc demoed a companion community tool called the Rayfin Fabricator, which strips away the CLI friction and lets developers go from a Power BI semantic model to a working, interactive application in a matter of minutes.
This isn’t another low-code toy. It’s Microsoft closing the gap between “we vibe-coded a prototype” and “this is running in production, governed, inside our data estate.”
What Rayfin Actually Is
Under the hood, a Rayfin app can be backed by two different data layers:
- A Fabric SQL database — for operational data: forms, to-do lists, transactional records, anything an app needs to write and manage.
- A Power BI semantic model — for apps that need to surface analytics and BI on top of data you already have modeled.
When you deploy with the CLI (rayfin up), Fabric provisions everything the app needs automatically: database, authentication, access policies, and APIs — no manual infrastructure work. Once running, the app’s data lands directly in OneLake, instantly available to the rest of the Fabric stack: Power BI for reporting, notebooks for data science, data agents for AI workflows. No pipelines, no data copying, no separate analytics environment.
The governance story is the part worth underlining. A Rayfin app is a Fabric item — identity is brokered through Entra, access is governed by the same workspace permissions that already protect your lakehouses and semantic models. That means when someone on your team ships an app this way, it isn’t shadow IT hiding in a corner. It inherits the same compliance boundary as everything else you own in Fabric.
The Tutorial: Idea to App in Minutes
Here’s roughly what the workflow looks like, as demoed with the Fabricator tool:
- Prerequisites. You need a working Power BI semantic model and the Rayfin SDK (or the Fabricator, which wraps it).
- Install the Fabricator. It scaffolds a new project directly from your semantic model, handling configuration and dependencies so you’re not managing terminal commands by hand.
- Generate with natural language. Instead of manually laying out a report canvas, you prompt the tool: add a page, add a slicer, filter this visual by region. The Fabricator reads your semantic model to keep the generated UI consistent with your existing data logic. It even supports different GitHub Copilot modes depending on how much AI assistance you want.
- Refine by hand when needed. The generation gets you most of the way there fast, but you’re not locked in — you can drop into VS Code and edit the underlying code directly, or use the built-in design mode to reposition elements manually.
- Deploy. One command hands you a live, authenticated, governed URL running inside your Fabric capacity.
The whole loop — idea, prompt, working app — happens in minutes, not sprints. That’s the headline. But it’s worth saying plainly: this is still in preview. Treat it as a pilot candidate, not a production migration plan, until the governance, CI/CD integration, and testing story matures further.
The Strategic Read: Does This Replace Power BI?
No — and that’s the more interesting point. Fabric Apps aren’t a Power BI killer; they’re an acknowledgment that “dashboard” and “application” have always been two different jobs wearing the same platform’s clothes.
Power BI’s report canvas is optimized for exploration and storytelling around a semantic model. It’s excellent at that. But it was never really built for interactive, write-back, app-like experiences — task flows, forms, operational workflows layered on top of analytical data. Translytical task flows and User Data Functions have been chipping away at that gap for a while; Rayfin is Microsoft going all-in on it with a proper application layer.
The practical implication for data teams: the center of gravity is shifting from report building to data product engineering. The semantic model becomes the reusable asset that feeds both a Power BI report and a Rayfin app, rather than being locked inside one report’s PBIX file. Teams that have been investing in clean, well-governed semantic models are the ones positioned to benefit fastest — the modeling work isn’t wasted, it’s now doing double duty.
What doesn’t change: the skills that actually make data work valuable. Storytelling, data modeling discipline, and a real understanding of what the user needs — those remain the hard, human part. AI is compressing the distance between “idea” and “working interface,” not replacing judgment about what to build and why.
Why This Matters Right Now
For any team currently framed as “the Power BI group” or “the reporting team,” this is a useful external data point. It’s evidence that the market — not just internal ambition — is already moving reporting-focused teams toward a broader “data product” mandate: owning the semantic layer as reusable infrastructure that powers reports, apps, and AI agents alike, rather than owning a set of dashboards.
It’s early. It’s preview. But it’s the direction of travel, and it’s worth having an opinion on before someone else forms one for you.
Rayfin is open source and Fabric-native by default, though self-hosting is technically possible given its code-first design. Expect the tooling — and the governance conversation around it — to keep evolving through 2026.
#MicrosoftFabric #Rayfin #PowerBI #FabricApps #MSBuild2026 #DataGovernance #AI #GitHubCopilot #DataEngineering #VibeCoding
MasterDataDNA 
Leave a comment