Fabric Weekly

Issue 13

6th October 2026

📣 Welcome to issue #13 of Fabric Weekly!

Microsoft's FabCon Europe announcements landed last week (catch up in issue #12), so this week it's the community's turn: recaps, first impressions and some useful warnings...

A few great FabCon recaps (useful for sifting through the deluge of announcements!):

  • Microsoft Fabric: Microsoft Fabcon Europe 2026 Announcements! from Bradley Ball, Josh Luedeman and Bradley Schacht - a keynote-day livestream with a view from inside the Fabric product group, covering the reworked ontology, the new Fabric policies and on-demand billing, which comes with per-workload spend limits.
  • Fabric on-demand capacity: the first experiences from Reitse Eskens - early private-preview impressions of F0 and on-demand billing, which look like a great fit for intermittent dev and light nightly workloads. Pricing hasn't been published yet, so don't plan around it just yet.
  • Microsoft Fabric Just Got a MASSIVE FabCon Update from Aleksi Partanen - a walkthrough of the September feature summary with Heini Ilmarinen, Oskari Heikkinen and Teemu Multanen, including where they'd be careful.

A couple of deeper looks at the new Deployment Plans:

A change in default settings that might be worth checking:

And two, more general, posts about how to set up good foundations in Fabric:

Finally, some UK events in the next couple of weeks:

  • Wednesday 7 October, Birmingham - #PBIBRUM with Victor Rivas and Jerome Wittersheim.
  • Tuesday 13 October, Manchester - PBIMCR at Data Decoded with Rob Sewell and Duncan Boyne (free Data Decoded ticket needed).
  • Tuesday 13 October, online - Devon & Cornwall Power BI with Cristi Prifti on MCPs, Skills and Plugins.
  • Thursday 15 October, Birmingham - FabFest, the free community Fabric conference.
  • Thursday 15 October, Bristol - Fabric & Power BI Bristol with Johnny Winter and Neil Rowlands (Meetup currently shows a different date, so check before travelling).
  • Thursday 15 October, Norwich - Norfolk Power Platform User Group with Simon Owen and Emily Delva.

🤖 Copilot, AI & Agents

  • Programmatically Control Copilot access for semantic models by Sandeep Pawar · 1 min read A new semantic model setting announced at FabCon Barcelona controls whether read-only users can use a model in Copilot, report Copilot and data agents; it is on by default, and anyone with write permission can toggle it. Sandeep Pawar shows how to read and set it in bulk from a Fabric notebook with SemPy, using an internal API that may change.
  • Fabric-RLM : Working With Semantic Models by Sandeep Pawar · 3 min read Sandeep Pawar shows how to bind Power BI semantic models into Fabric-RLM to answer questions and generate business artifacts in a Fabric notebook: the run inspects the schema, writes and self-corrects DAX, and records which measures each number came from. A worked example produces an HTML report on the biggest category revenue change, with custom checks that recompute key figures so a wrong report is rejected.
  • Prep data for AI vs. Fabric Data Agent Instructions: What Goes Where? by James Serra · 8 min read The final part of James Serra's three-part series on AI-ready data explains what belongs where when a Fabric Data Agent uses a Power BI semantic model: agent instructions handle routing, cross-source behaviour and response format, while Prep data for AI (AI data schemas, AI instructions and Verified Answers) holds the model-specific context, because the DAX-generation tool never receives the agent instructions. It also covers what differs for SQL, KQL and graph sources, and recommends testing against known answers by inspecting the generated query.
  • Microsoft Fabric: Microsoft Fabcon Europe 2026 Announcements! by Bradley Ball, Josh Luedeman, Brad Schacht · 1 min read Bradley Ball, Josh Luedeman and product group guest Bradley Schacht (with Christopher Schmidt dialling in from Barcelona) react to the FabCon Europe 2026 keynote live: the reworked ontology, which no longer builds a graph by default and can act as a data agent context source with per-source query instructions; Fabric policies for controlling which items can be created in a workspace, and a lively debate about whether they add yet another permission layer; F0 capacities with on-demand billing (warehouse first) and per-workload spend limits; GPU-accelerated warehouse in limited preview in selected regions; plus deployment plans (cautiously optimistic), Git branch workspaces with selective branching, the OneLake catalog search API and new Fabric apps features.

📊 Reporting & Insights

  • Microsoft Fabric Just Got a MASSIVE FabCon Update by Aleksi Partanen Aleksi Partanen walks through the most interesting of the roughly 150 announcements in the September Fabric feature summary with Heini Ilmarinen, Oskari Heikkinen and Teemu Multanen: app building in Power BI Desktop and Fabric Apps for Pro, semantic views, policies in Fabric, deployment plans, Project Osmos, Fabric IQ MCP and ontology updates, pipeline-level dependencies, and the upcoming F0 capacity with on-demand billing, with views on what they'd use and where to be careful.
  • Power BI Architecture Choices for Azure Databricks (w/ Liping Huang) by Reid Havens · 1 min read Reid Havens' livestream on 16 October (9:30 AM Pacific) with Liping Huang, author of Microsoft's white paper benchmarking four ways to report on Azure Databricks data in Power BI: Direct Lake on OneLake, Direct Lake over mirrored Unity Catalog tables, DirectQuery and composite models with aggregations. No single pattern wins: Direct Lake is best for warm and hot cache, aggregations stay under 100ms, DirectQuery catches up at 2.6 billion rows, and mirrored Unity Catalog tables slow down badly as data grows.
  • FabCon Europe 2026: 4 announcements that may change how we work with data and AI by Nikola Ilic · 8 min read Nikola Ilic picks the four FabCon Europe 2026 announcements he thinks matter most: the zero-provisioned F0 SKU with on-demand billing extended to every workload (pricing not yet published), apps in Power BI coming to Pro and PPU in preview with up to 1 GB of Fabric Database per app, Fabric IQ in Copilot Chat and Cowork now generally available alongside the Fabric IQ MCP server, and a data engineering agent in preview plus a three-step ADF and Synapse upgrade path. He warns that per-model AI access for read-only users is on by default, and suggests preparing semantic models and gathering capacity metrics now.
  • FabCon Europe 2026 Recap by Valerie Junk · 1 min read Valerie Junk recaps FabCon Europe 2026 in Barcelona: appearing in the keynote's community video, joining the Microsoft Tech Collective of content creators, demoing a Fabric Apps inventory and forecasting app that replaces everyday spreadsheet and email workarounds on top of an existing semantic model, and mornings at the community lounge. Her picks from the news are app building in Power BI Desktop and the F0 capacity with on-demand billing, coming in preview with pricing not yet shared.
  • FabCon & SQLCon 2026: The Ultimate Roundup of Announcements by Pragati Jain · 4 min read Pragati Jain rounds up the FabCon and SQLCon Barcelona announcements, picking the F0 zero-provisioned SKU with on-demand billing as her favourite, alongside Fabric IQ in Copilot (generally available), app building in Power BI Desktop for Pro and PPU users, Fabric Apps with the Rayfin SDK, IQ sharing (preview), the data engineering agent (preview), and SQLCon's Database Hub and DiskANN vector indexes, each linked to the relevant official post.
  • The Problem With Approximate Distinct Counts In Power BI – And How To Solve It by Chris Webb · 2 min read Chris Webb tests ApproximateDistinctCount(), newly supported on Import and Direct Lake models in the September 2026 Power BI release, on a 1.4-billion-row Direct Lake model: a query dropped from about 6 to 4 seconds and CPU time from 82 to 48 seconds, though memory use rose and results were about 1.6% off. To win over users who demand exact numbers, he suggests a field parameter that lets them switch between approximate and exact distinct counts.
  • Vibe-coded apps with Microsoft SSO out of the box by Aleksi Partanen Roey Zalta explains that Fabric Apps (built on Rayfin, in preview) host an app's back end, front end and database inside Fabric with Microsoft SSO out of the box, so only people granted access in Fabric can sign in. A clip from Aleksi Partanen's Fabric ontology conversation with Roey; the full video is linked.
  • Microsoft Fabric and SQL Community Conference 2026, the Barcelona edition by Reitse Eskens · 5 min read Reitse Eskens recaps his FabCon and SQLCon week in Barcelona: keynote highlights including on-demand (F0) capacity, Fabric policies, auto-pause for Azure SQL Hyperscale serverless in preview, observability and AI-assisted troubleshooting; standout sessions on Fabric Runtime 2.0 and the Native Execution Engine, with Runtime 1.3 heading for end of support; and his ambivalence about AI alongside a still-unsolved item ownership gap.
  • Activator and Power Automate – Turning Fabric Signals into Action by Laura Graham-Brown · 1 min read Laura Graham-Brown shows how to turn a Fabric signal into action: an eventstream of Microsoft's sample bike-share data feeds an Activator rule (such as more than 30 bikes at a location) that triggers a Power Automate flow.
  • Direct Lake Performance Based on Delta Layout by Vojtěch Šíma · 1 min read Vojtěch Šíma benchmarks 64 semantic models on an F32 to show how Delta layout drives Direct Lake performance and cost. Row group count dominates: delta-rs defaults warmed up fastest but cost 1.63x the readHeavyForPBI baseline in CU-s, while optimised V-Order layouts saved roughly 18–25%. His quick advice is Spark with V-Order and the readHeavyForPBI profile; for fine-tuning, tune row groups first, remembering that faster can mean noticeably more expensive.
  • Broken DAX? Asking twice used to be the fix by Aleksi Partanen Advanced DAX generation for data agents (in preview) now iterates and validates its DAX before responding, replacing the old workaround of asking the agent to fix a failed first attempt. A clip from Aleksi Partanen's August Fabric feature summary with Oskari Heikkinen.

⚙️ Data Engineering

  • Cosmos DB Mirroring – NoSQL in Fabric by Matthias Falland · 1 min read Matthias Falland's Fabric Friday episode explains that Cosmos DB mirroring runs on continuous backup rather than the change feed, which is why it consumes no request units but inherits rules such as no multi-region writes. He builds a mirror live, separates the snapshot and stream "clocks", and walks through the traps: a discovered schema that only grows, renames that silently leave nulls, nested data arriving as a string, and stopping replication not being a pause.
  • Does all your data have to sit in OneLake? by Aleksi Partanen Mark Kromer says your data doesn't all have to sit in OneLake: Data Factory pipelines, copy jobs and dataflows can move and transform data across Snowflake and Databricks too, even if Microsoft would like you to centralise. Another clip from Aleksi Partanen's conversation with Mark; the full video is linked.
  • Greenfield project? Skip mapping data flows by Aleksi Partanen Data Factory product manager Mark Kromer agrees that new Fabric projects should skip mapping data flows, which were carried over from Azure Data Factory so existing ones keep working, and use Dataflow Gen2, dbt, Airflow or Spark notebooks instead. A clip from Aleksi Partanen's conversation with Mark; the full video is linked.
  • Identity in Fabric doesn't count in order by Aleksi Partanen Identity columns in Fabric Data Warehouse are allocated in batches for performance: values are guaranteed unique but not sequential, and gaps can occur, so don't rely on counting rows from the maximum identity value. A clip from Aleksi Partanen's July Fabric feature summary with Vesa Tikkanen.

🏛️ Storage & Platform

  • Why build on Power BI? A Microsoft PM explains by Aleksi Partanen Data Factory product manager Mark Kromer explains why Fabric was built on Power BI's SaaS model rather than as an Azure-style service: avoiding resource groups, control planes and ARM templates makes it less of a pro-developer experience and helps projects move faster. A clip from Aleksi Partanen's conversation with Mark; the full video is linked.

🔐 Governance & Security

  • Workspace Deep Dive – Four Boundaries in One Object by Matthias Falland Matthias Falland's Fabric Friday episode explains why creating a workspace sets four boundaries at once (security, cost, region and lifecycle), with demos showing how a Viewer can get results through the SQL endpoint but a 403 in Spark, why OneLake security can't fence in a Contributor, and who can assign a workspace to a capacity. A cross-region move cancels jobs, and only seven item types survive it.
  • #6 Fabric Tenant Settings: What to Enable, Disable, and Ignore at Launch by Paul Turley · 1 min read Part 6 of Paul Turley's Fabric & Power BI Essentials series looks beyond the headline tenant settings: keep guest-sharing sub-settings and Publish to Web off without a real business case, treat each export format as a separate decision, extend Purview sensitivity labels early, leave service principal developer settings off until a pipeline needs them, and defer tenant-level Private Link, public internet blocking and customer-managed keys until security has defined requirements.
  • 10 Decisions to Make Before You Create A Fabric Workspace by Meagan Longoria · 10 min read Meagan Longoria sets out ten design decisions to settle before creating Fabric workspaces, since restructuring later means redeploying content and rebinding references: capacity and region, network security posture (including what tenant-level Private Link disables), content ownership, how to split workspaces, environments, source control and deployment, naming conventions, domains, access and distribution, and classification with sensitivity labels and tags. Each comes with the options, what should drive the choice, and the platform constraints that narrow it.

💰 Management & Cost

  • Enrich your Fabric CI/CD experience with deployment plans by Kevin Chant · 1 min read Kevin Chant explains how the new deployment plan item (preview) enriches Git integration, deployment pipelines and REST-based deployments rather than replacing them, with three uses: resolving dependency order (deploy and populate a Lakehouse before a warehouse view that references it), DataOps via post-deploy notebooks when branching out or promoting through deployment pipelines, and rebinding items such as semantic models with a customised version of Microsoft's branch-out notebook. fabric-cicd doesn't support plans yet.
  • Fabric on-demand capacity: the first experiences by Reitse Eskens · 2 min read Reitse Eskens shares early impressions of Fabric's new on-demand capacity from private-preview access, ahead of official documentation: no compute charge when idle (storage is still billed), a configurable maximum CU cap, no reservations yet, and some items expected to be priced at two or three times regular capacity rates. He sees it as a strong fit for intermittent development and light nightly ETL workloads, with reserved dedicated capacity still making more sense for sustained use.
  • Fabric Deployment Plans - Swipe Left or Swipe Right by Jake Duddy · 1 min read Jake Duddy unpacks the new deployment plan item (preview): a git-committed plan.yml that adds explicit deployment order and pre- or post-deploy actions (notebooks, pipelines, Dataflow Gen2, copy jobs, user data functions) on top of Fabric's lineage ordering, attached per operation to Git updates, branch-out, deployment pipelines or REST calls. He swipes left for now: fabric-cicd doesn't support plans, auto-binding gaps remain, variable libraries read the Default value set in new workspaces, plans are single-workspace, and the first failure halts the deployment with no rollback.
  • Fabric CICD - Two Different Worlds by Jake Duddy · 2 min read Following his Deployment Plans post, Jake Duddy argues that Fabric CI/CD has split into a low-code portal world (git integration, deployment pipelines, Variable Libraries, deployment plans) and a code-first repo world (fabric-cicd, parameter.yml, YAML pipelines) that don't meet. He catalogues where Microsoft's documentation contradicts itself within each, lists the binding, empty-workspace, service principal and preview gaps, and asks for logicalIds on every reference and on workspaces, plus feature parity so the two worlds become two front ends to one platform.
  • Fabric-CICD in Practice - Groundwork by Jake Duddy · 1 min read Part 1 of Jake Duddy's new series building fabric-cicd deployment into empty workspaces with no manual clicks: Release Flow branching mapped to feature, dev, test and prod workspaces, two service principals using workload identity federation (no secrets), least-privilege tenant settings, config.yml and parameter.yml, and two Azure DevOps pipelines with an approval gate on prod, ending with a cold-start deployment into empty test and prod workspaces.
  • Your Fabric capacity is throttling. Don't double it (yet) by Dániel Gábor Patkós · 1 min read Part 2 of Daniel Patkos's Fabric capacity myths series argues that throttling shouldn't trigger an automatic SKU doubling: investigate with the Capacity Metrics App first, then consider targeted fixes such as splitting workloads across capacities, Pro workspaces for reporting, a paused "rescue" capacity, scheduled scaling or a short-lived PayGo capacity for one heavy refresh. He also warns that pausing settles smoothed usage immediately, so it doesn't always save money.

📅 Community & Events

Get Fabric Weekly in your inbox every Tuesday

Sign up to receive the email every week and keep on top of all the announcements.

Don't forget to confirm your subscription. We have a double opt-in sign up process to prevent spam.

We will never provide your details to any third parties. We don't spam.

Fabric Weekly Logo