On the official side:

• The headline is GPU-accelerated query execution for Fabric Data Warehouse - an informal-but-rigorous TPC-H benchmark showing roughly 6x the throughput at 1000 concurrent users versus CPU execution, with p95 latency dropping from 97 seconds to 15.
• Also worth a look: role-based access control for API for GraphQL lands in preview, with field- and row-level security baked into the API itself rather than a custom app layer.
• OneLake picked up a genuinely money-saving GA feature: lifecycle management with storage tiers, which automatically moves cold data into cheaper cool/cold tiers based on access patterns without you lifting a finger.
• And Fabric Data Factory pipelines got a batch of new capabilities, headlined by an AI-powered Operations Agent (preview) that watches pipeline health and suggests fixes, plus activity-level monitoring so you can drill straight from a failed run into the specific step that broke.

Two bigger stories worth their own mention this issue: Reza Rad sat down with Yitzhak Kesselman (CVP for Real-Time Intelligence at Microsoft Fabric) for a expansive conversation on where Fabric IQ and Real-Time Intelligence are heading - ontologies, ops agents, and the "no AI without RTI" argument, with real customer numbers to back it up. And Rayfin, aka Fabric Apps - Microsoft's new open-source SDK for building full-stack web apps entirely inside a Fabric workspace - TypeScript frontend, Fabric SQL database or Power BI semantic model backend, all under one governed roof.

In the community:

• Erwin de Kreuk's FMD framework can now generate column-level lineage from Fabric Lakehouses straight into Microsoft Purview.
• And a deep, honest write-up on getting governed Databricks data into Fabric behind a fully private network compares four setups and lands on the same conclusion every time: the hard part isn't moving the data, it's carrying the Unity Catalog policies with it.
• And Chris Webb finally sets out the actual benefits of Direct Lake mode beyond "it's the new shiny thing" - cheaper refresh, no memory limits, and reuse of the same OneLake tables across multiple semantic models.

And, over on the endjin blog: a blog from yours truly which compares star schema, flat table, and header/detail designs in DAX Studio.

On the official side: the headline is the Fabric data agent API going public - the data agent SDK now runs on the Fabric REST API instead of being stuck inside a notebook, so you can create, configure, and publish data agents from your own CI/CD pipelines, internal tools, or backend services. Big news for anyone provisioning the same agent shape across dozens of workspaces.

Also worth a look: Delegated OneLake Shortcuts hit preview, letting a shortcut route access through a service principal instead of every consumer's own identity- This solves both the "central team can't scale permissions" problem and cross-tenant sharing. And Item Recovery is now GA, so "I accidentally deleted the wrong lakehouse" is a recoverable mistake!

Over in the community: Reza Rad shows how pairing GitHub Copilot with Microsoft's open-source Fabric Skills library turns "list every empty workspace in my tenant" from a manual job into a single prompt.

Bart Wullems shows how Fabric's Copy job now builds Slowly Changing Dimension Type 2 for you with zero merge statements - provided your source/destination pair is currently supported. And on the reporting side, Marco Russo and Alberto Ferrari debug a running-total measure that quietly broke because of sort-by-column settings, fixing it properly by packaging REMOVEFILTERS into a reusable user-defined function instead of patching the one measure.

Last but not least, Kevin Chant shows how Managed DevOps Pools can get your Fabric deployment pipelines into a firewall-protected Key Vault without the usual self-hosted-agent headache.

On the official side:

The headline is Fabric Data Agents now supporting Microsoft Entra ID service principal authentication - one of the most requested capabilities for moving data agents from prototype to production, and the start of a short video series if you want to follow the setup step by step.

Also worth a look: Event-Driven Copy Job Execution with Fabric Activator is now GA, letting you trigger Copy jobs straight from an Activator rule with no pipeline in between, and there's a new Fabric Spark Operations Skill (Preview) that turns "why did my notebook fail at 3am" into a single natural-language prompt instead of a multi-tab log hunt.

Microsoft is also bringing Power Query connectors fully in-house with a clearer Preview-to-Retirement lifecycle.

And Build 2026 brought a long list of announcements covered in James Serra's roundup: Agent Skills for Power BI is the one most likely to change how you build reports, while Rayfin - the new open-source SDK that lets coding agents define and deploy a full governed app backend straight onto Fabric - might be the one that changes what counts as a Fabric workload at all.

Over in the community:

Roman Lánský dug into why your mirrored "real-time" reports might quietly be serving stale data, and what the new SQL endpoint metadata sync does about it.

Nikola has written the definitive (and very long) guide to variable libraries, the Fabric item that exists specifically so "Test is reading Prod" becomes a story you only tell at conferences.

John Kerski makes the case that DAX User Defined Functions are the biggest thing to happen to semantic model testing since PBIP, and has the downloads on his PQL.Assert library to back it up.

Meagan Longoria offers a sensible crawl-walk-run framework for agentic Power BI development if you'd rather not hand an AI agent the keys all at once

And, Adam Marczak argues it's time to stop treating Azure itself as the platform and pick Fabric or Databricks as the actual center instead.

Thanks for being here for issue #1 - tell a friend, tell your whole workspace...