Microsoft Fabric

Episode 324 – Microsoft Fabric March 2026 Feature Summary (Part 2)

Episode 324 picks up where Episode 323 left off and keeps the train rolling through the Microsoft Fabric March 2026 Feature Summary. Recorded on Friday, April 17th, 2026 — with John cranking towards a Canadian spring and Jason heading into a weekend of theater for the dude — the hosts skip straight into the Data Science & AI section, plow through the Data Warehouse updates, then pull the Databases section ahead of Real-Time Intelligence to give the new Database Hub the oxygen it deserves.

Data Science & AI (the rebrand nobody announced)

The section formerly known as “Copilot and AI” has quietly been renamed Data Science & AI, and Jason and John agree it’s the right call. Copilot is about using Fabric with AI; this section is about building AI constructs on top of Fabric. As Jason puts it, “it’s so much broader than just Copilot now between MCP servers and everything else.” Drawing a clean line between the two makes it easier to follow where the platform is heading.

Fabric Data Agents are GA

Fabric Data Agents have gone generally available. John’s take is that the announcement is more about the lifecycle scaffolding — provisioning, Git integration, diagnostic and deployment pipelines, troubleshooting — than it is about net-new behavior, but those are exactly the boxes any item has to check to earn GA in Fabric. Under the preview tag, data agents also picked up advanced security and governance via Purview integration, and data source enhancements now let agents reach into graph items (Fabric’s graph data type, not Microsoft Graph), KQL user-defined functions, and SQL functions.

Jason is enthusiastic about the graph direction in particular — graph data is, by design, a natural fit for how agents reason — and he’s hopeful that over time data agents will reach further into the wider Microsoft Graph and Copilot / Foundry IQ surface. John’s dry response on that one: “I’ve been waiting for that for 15 years.”

Multimodal AI functions and AutoML

The blog post had a section-title hiccup — Jason read “Insider Risk Management pay-as-you-go reporting” out loud and John immediately shut that down because the hosts already covered IRM in Part 1. The actual topic is multimodal support for AI functions in Fabric. AI functions — the sentiment-analysis and translation primitives — now work against unstructured data via the files endpoint, not just structured data. Read a document, score it, translate it, return the result as data, all inside a notebook. That’s a real unlock.

AutoML in Fabric also reached GA this month. The functionality itself hasn’t fundamentally changed from its Power BI origins, but GA is GA — it’s now a supported part of the data science surface.

Data Warehouse: Recovery, Activator, Custom SQL Pools and more

Fabric Data Warehouse picked up a soft-delete / recycle-bin style recovery. Accidentally drop a warehouse, and it sits in a soft-delete state for up to 35 days where it can be restored. Jason flags the obvious caveat: if you meant to delete a warehouse because of a data leakage or compliance concern, you now need to explicitly purge it — otherwise it’s sitting there waiting to be brought back. A nice feature that administrators need to know about.

Activator now supports warehouse as a source — fire actions when query values cross thresholds, exactly the way you’d hope. It’s a good sign that Activator is broadening its scope; the hosts expect to see it show up in more parts of the platform over time. The T-SQL AI function is in preview, bringing the same AI primitives from notebooks into warehouse queries. Screenshots in the blog post show it inside a query editor, but because the engine is shared, Jason and John both suspect it works in a T-SQL notebook too — the blog just doesn’t call it out.

The ANY_VALUE aggregate arrived in Fabric Data Warehouse — a clean way to do GROUP BY-style queries without exploding on minor variance in non-key columns. Custom SQL Pools for Fabric Data Warehouse entered preview, and John has a sharp observation: “This is Fabric turning into Azure a little bit, right? You’re seeing this here and there moving a little bit more from a platform into infrastructure — knobs and buttons if you want to adjust them.” Administrators can now define their own SQL pools, allocate a percentage of backend compute per pool, route queries by application name, and get real workload isolation. Jason ties it back to conversations with performance-tuning DBAs who find fabric SQL lacking the tuning levers they’re used to — this starts to close that gap, and the hosts expect the same capability to land on Fabric SQL as SQL hyperscale rolls into Fabric down the road.

SQL audit logs reached GA — SQL analytics endpoint auditing, direct access to audit files in OneLake, download and copy through OneLake Explorer, SSMS for deeper investigation. COPY INTO and OPENROWSET support for OneLake sources also went GA, though the blog makes a point of calling out that warehouse is not a supported source here — a confusing placement that may have been a quiet heads-up in disguise. Outbound access protection for warehouse arrived this month too, following the same anti-exfiltration initiative that shows up elsewhere in the platform. Notably, this is a Fabric-native capability, not a Purview dependency — which Jason likes because it doesn’t force organizations into a Purview deployment to get basic data-leaving-the-tenant controls.

Query insights finally show the full query text — the 8,000-character cap is gone. It’s a bug fix dressed up as a feature — and fair enough. The live connectivity in the migration assistant is in preview, letting you actually connect to your source rather than running blind through DAC pack packages. And simplified data access with data sources went GA — Jason’s read on it is that it’s functionally a manual connection-string library with a nice declarative feel, which John doesn’t argue with. The pair have a moment of self-awareness: “A lot of what we do here is interpret what these weird titles mean.” Jason confirms listeners have said that’s exactly what they like about the show.

Databases: The Database Hub (the one Jason is pinning hopes on)

The hosts pulled Databases ahead of Real-Time Intelligence so they could give the Database Hub real airtime. Early access signup is linked directly in the blog post. Jason reads the pitch aloud: “The Database Hub in Fabric is a new unified database management experience that brings databases across edge, on-prem, cloud and Fabric into a single coherent view. It provides teams with one place to explore, observe, govern, and optimize their entire database estate.”

Then he pins his hopes on it. Azure SQL, on-prem SQL, Cosmos DB, Fabric SQL — one place to view logs, trace across systems, track everything, potentially orchestrate everything. The word “manage” doesn’t appear in the blurb, but Jason reads “optimize” as a wrapper around it, and John agrees that’s at least the vision. Microsoft has confirmed the hub will extend to non-Microsoft databases too, with mirroring-style connectivity back to source databases wherever they live. The hosts see the parallel to OneLake shortcuts — connect to databases where they live, provide visibility across the board — and both of them find it an interesting strategic choice to put this capability inside Fabric.

Jason lays out his expected rollout: Fabric SQL, Azure SQL and other Azure-hosted databases first; on-prem SQL second; other databases as the third wave. Microsoft’s incentive to drive people into Fabric — even as a management layer — is clear. He also makes a three-year call: this will be painful to live on for the first year, probably worth it by the end of year one, and a no-brainer by year three. Going straight into the prediction file for 2026. John notes that the companion SQL database updates — a redesigned migration assistant, an AI auto-tuning agent that scales vCores based on load, full collation support, mirroring with customer-managed keys, vector search, monitoring integration, and the same soft-delete recovery pattern — are going to pair nicely with all of this. Cosmos DB’s big headline: private link and VNet support. If that was blocking you, it isn’t anymore.

With that, Jason calls time. Real-Time Intelligence deserves its own block — it may even take a full episode — and Data Factory is still waiting. Episode 325 will pick up at RTI.

Links

Microsoft Blog Posts

Previous Episodes

Events

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

Tags: , , , , , , , , , , , , , , , , , , ,

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

AvePoint