Episode 293 – Microsoft Fabric March 2025 Feature Summary (Part 2)

May 7, 2025

John and Jason are back for Part 2 of their deep dive into the Microsoft Fabric March 2025 feature summary, and folks, there’s still a mountain of updates to unpack. Between birthday celebrations (Jason’s youngest turns 15!) and plumbing disasters, they managed to tackle data science improvements, warehouse enhancements, real-time intelligence updates, and Data Factory features that’ll make your development life easier.

Copilot Finally Gets Its Act Together

The biggest news buried in this episode? Copilot in notebooks is finally worth using. John’s verdict is crystal clear – if you tried it before and thought it was “a steaming pile of something or other,” it’s time to give it another shot. The old version required pre-loading libraries, took forever, and generally made you want to throw your laptop out the window.

Now? It works like an actual copilot experience with a proper pane on the right, code suggestions that don’t make you question your life choices, and agentic capabilities that might actually help instead of hinder.

But here’s the kicker – you need a paid SKU to access it. Jason discovered this the hard way when it worked beautifully in his customer’s environment but was nowhere to be found in their trial. The good news? It works perfectly fine on F2, so you don’t need to break the bank to get started.

AI Skills Gets a Rebrand (And More Power)

Microsoft decided that “AI Skills” wasn’t descriptive enough, so they’ve renamed it to “Fabric Data Agent.” Makes sense when you think about it – these things ground LLMs with your data and make them useful in places like Azure AI agent service.

The real magic here is the seamless integration between unstructured data (Azure AI search, SharePoint) and structured data in OneLake. Jason’s excited about this because it means all those agents he’s building in Azure and SharePoint can now tap into a much richer data pool. Plus, the new Python SDK for Fabric Data Agent gives developers programmatic control for fine-tuning.

Warehouse Gets Some Love (Finally)

John’s favorite topic gets some attention with AI functions now available in data warehouse and Lakehouse SQL endpoints. You can now summarize content, translate text from Romanian to English, extract key data, and do sentiment analysis – all directly in T-SQL.

As John put it, if you’re an ISV supporting multiple languages, you no longer need people manually inputting translations. The LLM tech will handle it, and you can fine-tune from there.

The warehouse also picked up some fit-and-finish improvements including proper expand/collapse in filters, better Git status bars showing branch names and sync status, and the ability to cancel queries when closing the editor instead of letting them run wild in the background.

Real-Time Intelligence Keeps Getting Better

John’s passion project continues to evolve with new sources for event streams including MQTT, Solace, Azure Data Explorer, and Azure Event Grid. But the real crowd-pleaser? A weather feed connector that lets you get actual real-time weather data for any location without buying your own weather station.

John and Jason have been running their own weather stations for seven years, solving the same problem with different tech stacks. Now anyone can get into real-time data without the hardware investment. John’s weather rock is apparently still faster than the official alerts, but that’s beside the point.

Event streams also reached general availability for CI/CD support and REST APIs, plus got managed private endpoints and Entra ID authentication for those organizations that refuse to touch anything on the public internet.

Data Factory Catches Up

The Data Factory section reads like a feature parity checklist, and that’s not a bad thing. VNET gateway support for data pipelines addresses security concerns, while variable libraries for pipelines (in preview) finally brings proper environment management to the platform.

The ability to save Data Flow Gen1 as Gen2 with CI/CD support is basically a “save as” operation that inserts what’s necessary to make it work. Think of it like saving a PPT file as PPTX – same concept, better format.

John’s particularly happy about the Apache Airflow job reaching general availability, though he admits he knows nothing about it beyond “if you don’t want to build a pipeline and you’re used to Apache Airflow, use it instead.”

The Bottom Line

This episode covers the remaining chunks of what Jason estimates would be 80-100 pages if printed. They’re catching up on Microsoft’s release cadence (currently a month behind with April already out), but there’s good reason for the backlog – the March release was absolutely massive thanks to everything announced at Fabric Conference.

What’s clear from this deep dive is that Microsoft is focused on making Fabric more enterprise-ready. Security features, CI/CD support, and better integration points are the common themes across every workload. The platform is maturing fast, and the gap between “cool demo” and “production ready” continues to shrink.

What’s Next

John and Jason are gearing up for a busy conference season with completely reworked content that takes advantage of new Fabric capabilities. They’re heading to Germany for ECS, then Seattle, Atlanta, Branson, and Dublin for ESPC with their tutorial and three new sessions covering CI/CD, real-time intelligence, and data transformation.

The March feature summary is finally complete, but April’s already waiting in the wings. With Microsoft’s release pace showing no signs of slowing down, they’ll need to stay on their toes to keep up.

---

Links:

• Episode on SoundCloud
• Microsoft Fabric March 2025 Feature Summary
• Episode 292 – March Feature Summary Part 1
• European Collaboration Summit

Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music