Fabric Weekly

Issue 14

13th October 2026

🍂 Welcome to issue #14 of Fabric Weekly!

The FabCon Europe dust is settling (catch up in issue #12 and issue #13), and this week a few more follow-up posts are landing, along with some additional announcements!

On the official side:

From the community:

One more FabCon roundup, and it's a thorough one:

A change in default settings that might be worth checking:

And a few practical reads:

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

🤖 Copilot, AI & Agents

  • A context layer for your entire company by Aleksi Partanen Roey explains how an ontology creates a company-wide context layer for AI agents in Microsoft Fabric’s IQ preview.
  • Column names are almost never descriptive by Aleksi Partanen The new schema object descriptions for Fabric data agents allow adding meaningful context to column names, improving answer accuracy and ensuring consistent metadata across SQL sources.
  • Fabric IQ in Microsoft 365 Copilot: Your Semantic Model Is Now the API by Matthias Falland Fabric IQ in Microsoft 365 Copilot converts the semantic model into an API enabling licensed Copilot users to receive data answers directly in chats and collaborative environments, impacting security, refresh policies, and supported content types.

📊 Reporting & Insights

  • ODBC to ADBC in Power BI and Fabric: Test Your Refresh Before Cutover by Matthias Falland Matthias Falland on the move of supported Power BI and Fabric connectors from embedded ODBC drivers to ADBC, now generally available for validation: test your most important Import and DirectQuery refreshes against today's ODBC baseline in a pilot workspace before cutover. It doesn't apply to the generic ODBC connector, and pinning Implementation 1.0 without a gateway will fail at cutover.
  • How I Architected a Complete Reporting Solution on Fabric F2 by Gilbert Quevauvilliers · 1 min read Gilbert Quevauvilliers describes a customer reporting solution running on a single reserved Fabric F2 ($182/month) plus an Entelexos Basic embedded portal ($180/month): 28 Python notebooks loading API data into a lakehouse, a metadata-driven pipeline running twice a day, and an Import semantic model serving interactive and paginated reports. He explains why app-owns-data embedding means viewers don't need individual Power BI licences, and when he'd reassess the capacity.
  • What’s new for Business Events and real-time events in Microsoft Fabric by Roberto Cervantes · 6 min read Business Events are now generally available in Real-Time Hub: governed, schema-versioned business signals published once from notebooks, User Data Functions or Activator and consumed independently through Activator, Eventhouse or Eventstream (the Eventstream consumer is in preview). The post also covers GA Capacity Overview Events, preview Capacity Operation Events for operation-level root cause, and GA workspace item events for soft-delete and recovery.
  • ApproximateDistinctCount In Power BI Is Sometimes Slower Than DistinctCount by Chris Webb · 2 min read Following last week's post, Chris Webb shows that ApproximateDistinctCount can be slower and use more memory than DistinctCount on low-cardinality columns, because some DistinctCount engine optimisations don't apply to it. In his Direct Lake test on a two-value column the approximate measure took 688ms against 188ms, so test before switching; it earns its keep on columns with millions of distinct values.
  • From D365 Business Central to Power BI: Transform Your Sales Data into Insights by Soheil Bakhshi · 1 min read Soheil Bakhshi walks through building a Power BI sales model on Dynamics 365 Business Central from scratch: connecting through Standard APIs v2.0, choosing four tables, cleaning them in Power Query, adding a date table, creating 50 measures at once in DAX Query View, and using Copilot to draft the report pages, which still need reviewing. The post accompanies a video.
  • Announcements from the Microsoft Fabric Community Conference — Barcelona 2026 by James Serra · 35 min read James Serra's extensive roundup of FabCon and SQLCon Barcelona 2026, grouping the announcements by status (GA, public preview or coming soon, private preview or sneak peek) and then by area, and noting where keynote and documentation labels differ. His favourites include Fabric IQ in Copilot Chat and Cowork, IQ Sharing, agentic data engineering with Osmos, on-demand billing and F0 capacity, Policies in Fabric, the new unified Copilot and Modern Report Builder.
  • Planning in Fabric IQ: Budgets and Forecasts on Your Semantic Model by Matthias Falland Matthias Falland on Plan in Fabric IQ (GA since July 2026), which puts budgets, forecasts and scenarios on the same semantic models as reporting. It suits finance teams already on governed models on Fabric or Premium capacity; limits include no Entra B2B users, no plan items in private-link workspaces, no Pro or PPU for XMLA-based planning, and write-back of up to 1.2 million cells per operation.
  • Fabric Apps Hands-on: From Empty Folder to a Running App with Rayfin by Matthias Falland Fabric Apps enables developers to swiftly create a running application from an empty folder using two CLI commands within a Fabric workspace preview, suitable for internal tools but lacking private link support, outbound access protection, and app lifecycle management.
  • Fabric Apps vs Power Apps vs Azure: When to Build an App on Fabric by Matthias Falland Fabric Apps can now query lakehouses, warehouses and semantic models directly and run server-side code (preview). Matthias Falland compares them with Power Apps and classic Azure apps: build on Fabric for developer-built internal tools or agent backends when the data is already in Fabric, choose Power Apps for low-code makers, and Azure if you need private links, outbound access protection or ALM.
  • Fabric Business Events: Publish Once, Let Any Team Subscribe by Matthias Falland Business Events in Fabric Real-Time Intelligence are now GA: publish a governed business signal once and let any team subscribe without changing the producer. Matthias Falland explains when they fit (not for raw telemetry or logs) and the running costs: per-hour event listeners, every delivery attempt including retries, and a default eventhouse that keeps every event unless you untick it.
  • Understanding DAX contracts when using AI by Alberto Ferrari · 1 min read In this video, Alberto Ferrari introduces defining a 'contract' as a key activity for DAX developers working with AI, illustrated with simple examples.
  • Where Writeback Belongs – Ep.569 by Mike Carlo, Tommy Puglia · 1 min read In this episode, Tommy Puglia and Mike Carlo answer a mailbag question on whether Power BI users should be able to act on a number without leaving the report, arguing that writeback belongs in Fabric apps, the report stays for consumption, and translytical task flows are the exception.
  • Code Coverage for Semantic Models: Introducing pql-test code-coverage by John Kerski · 4 min read John Kerski introduces code coverage for semantic models in pql-test 0.1.18. A table, column, measure, relationship or role counts as covered only when a DAX test UDF references it directly; the report lists gaps and can fail a run below a minimum threshold. He's clear that coverage shows an object is referenced by a test, not that its behaviour is validated.

⚙️ Data Engineering

  • Custom SQL Pools in Fabric Data Warehouse (Generally Available) by Twinkle Cyril · 2 min read Custom SQL Pools in Fabric Data Warehouse are now generally available, letting admins isolate workloads such as ingestion, reporting and ad hoc analysis into pools with maximum resource percentages. New statement-type classification splits SELECT and non-SELECT requests into separate pools (50% each by default, adjustable), alongside the existing classification by application name or regex.
  • dbt Job in Microsoft Fabric: Where Should SQL Transformations Live? by Matthias Falland The integration lets Fabric users author, schedule, and monitor dbt projects with serverless execution, ideal for SQL-first teams using Fabric targets, but not recommended for very large projects lacking build caching or when existing logic is already split across multiple tools.
  • Fabric Data Engineering Agent: Hand Over a Pipeline, Keep the Keys by Matthias Falland Matthias Falland on the Fabric data engineering agent (Project Osmos), now in preview: describe an outcome such as a medallion build or a move to incremental loads and it plans, builds, runs and validates the work in Fabric. It isn't meant for quick queries or production yet, and it acts with your permissions even outside the workspace, so run it under a narrowly scoped identity; it can't be used in workspaces with outbound access protection.
  • Materialized Lake Views in Fabric: Medallion Without the Notebooks? by Matthias Falland Materialized lake views are now GA: declare bronze-to-gold transformations in Spark SQL and Fabric runs them in dependency order. Matthias Falland covers where they fit (repeated aggregations, joins, quality rules) and where they don't (ML inference, API calls, complex Python, sub-second streaming), plus the fine print: window and non-deterministic functions, and updates or deletes without a refresh hint, fall back to a full refresh.
  • Copy Job CDC and SCD Type 2 in Fabric: History Without Custom Code by Matthias Falland The September 2026 Fabric update generally available adds CDC and SCD Type 2 to Copy jobs, enabling automatic history maintenance for slowly changing dimensions without custom code, though it has limitations with complex business rules or Oracle sources.

🏛️ Storage & Platform

  • Microsoft Fabric Domains Explained: Roles, Default Domain, Labels by Matthias Falland Microsoft Fabric domains act as catalog filters and metadata attributes for governance, not security boundaries, allowing every tenant user to see all domain names while delegating settings and default sensitivity labels.
  • Microsoft Fabric: A domain is a label, not a lock #Shorts by Matthias Falland In Microsoft Fabric, domains function solely as labels indicating ownership, while workspace roles and item permissions determine actual access rights.
  • Fabric Mirroring Change Data Feed: When Free Mirroring Starts to Cost by Matthias Falland Change data feed for mirroring is now GA, and it's the first paid feature inside mirroring itself. Matthias Falland explains it's worth enabling only on mirrored databases with a genuinely incremental consumer (a silver-layer merge, delete auditing, a stream into Eventhouse), since it's billed per mirrored database rather than per table, and the change files add storage.

🔐 Governance & Security

  • Who uses customer managed keys? I asked the panel by Aleksi Partanen A clip from Aleksi Partanen's July 2026 feature-summary panel, in which Vesa Tikkanen, Oskari Heikkinen and Heini Ilmarinen discuss how often their clients use customer-managed keys in Fabric: mainly regulated finance and public-sector customers.
  • Fabric Policies: Who May Create What, Decided per Capacity by Matthias Falland Matthias Falland on Fabric policies, now in preview: centrally defined governance rules at tenant or capacity scope, with conditions such as workspace, security group or sensitivity label. Three policies are available today (item creation per capacity, external data sharing, workspace settings editing), and they aren't yet supported in the West Europe, North Europe or West US capacity regions.
  • OneLake Outbound Access Protection: Which Way Does Your Data Leave? by Matthias Falland OneLake outbound access protection is now GA: in a protected workspace, external shortcuts and copy operations are blocked by default and only approved connections are allowed. Matthias Falland covers when it's worth it and the trade-offs, including incompatibility with external data sharing, no cross-tenant allow lists, and limits on Power BI items in protected workspaces.
  • Fabric Core MCP Server: What an AI Agent Can Do in Your Tenant by Matthias Falland The new Fabric Core MCP Server enables secure AI agent management of Fabric resources with authenticated access to workspaces, items, permissions, and search data, requiring careful permission settings and client configuration.
  • #7 Planning Your Fabric Rollout: Pilot vs. Enterprise-Wide Adoption Strategies by Paul Turley · 1 min read Part 7 of Paul Turley's Fabric & Power BI Essentials series argues the real question is which approach your organisation is ready for: an isolated proof of concept, then a pilot with roughly 10-15% of eventual users to surface governance and skills gaps, or a broader rollout from day one if you already have a mature Power BI footprint and an engaged sponsor. Define success and exit criteria up front, and add workloads only as maturity allows.
  • SQL security won't stop Spark users by Aleksi Partanen Aleksi Partanen on why SQL analytics endpoint security doesn't stop Spark users: anyone with access to the lakehouse can read the same data from a notebook. OneLake security is what governs readers, and the endpoint only applies it in user's identity access mode.

💰 Management & Cost

  • A new look and new capabilities in the Capacity Metrics App (Preview) by Pankaj Arora · 3 min read Announced at FabCon Barcelona, a preview version of the Capacity Metrics app now installs alongside the existing GA version. It adds a health page that compares capacities by measures such as percentage of time spent throttled, a compute page with CU-second utilisation, an hourly heatmap and a smoothed CU consumption table, a refreshed storage page, and a new item details page with 14 days of history.
  • Fabric Capacity Metrics App, a huge upgrade! by Reitse Eskens · 2 min read Reitse Eskens takes a first look at the new preview Capacity Metrics app, which installs as a second option alongside the existing one, walking through the colour-coded health overview, the reworked compute graph with its hourly heatmap, the improved storage view and timepoint detail, and urging admins to install it and send Microsoft feedback.
  • Official icon sources for Microsoft Fabric architecture diagrams by Kevin Chant · 1 min read Kevin Chant shares where to find the official icons for Microsoft Fabric architecture diagrams, including related services such as Azure DevOps, Git and GitHub.
  • Microsoft Fabric Deployment Plan Explained: Run Code During Deployments by Aleksi Partanen Aleksi Partanen and Teemu Multanen walk through the new Deployment Plan (preview): actions that run notebooks or pipelines before or after each item, dependencies, Variable library parameters, the plan as YAML in Git, and running it from deployment pipelines and branch-out. Two gotchas: selecting only the plan deploys nothing else, and a failed action stops the deployment with no rollback.
  • Fabric Branch Workspaces: One Workspace per Feature, Without the Mess by Matthias Falland Fabric’s new branch workspaces simplify parallel development by isolating each feature in one step, though they require capacity planning and careful management to prevent orphaned environments.
  • Fabric Capacity Overage: Pay Instead of Throttle, and When Not To by Matthias Falland Capacity overage is now GA for F SKUs and is switched on by default for new capacities, billing temporary overloads instead of throttling. Matthias Falland explains when that's worth it, why daily throttling is better solved by scaling up (keep the threshold below about a third of daily CU hours), and how it works alongside surge protection.

📅 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