Fabric Weekly

Issue 17

3rd November 2026

🎆 Welcome to issue #17 of Fabric Weekly!

A good week this week: an agent that takes on whole data engineering tasks for you, a couple of nice GAs, and plenty of great community deep dives.

This week, on the official side:

From the community:

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

🤖 Copilot, AI & Agents

  • Microsoft 365 Copilot And Power BI: The Easiest Way To Chat With Your Data by Chris Webb · 5 min read Chris Webb's advice for Power BI shops wanting to let users chat with their data: start with the now-GA integration between M365 Copilot (and Cowork) and Power BI semantic models, rather than Data Agents or ontologies. It reuses your existing AI prep, can draw on Work IQ context, and with an M365 Copilot Business Premium licence the only extra cost is the CUs for the DAX queries.
  • Introducing the data engineering agent for Fabric (Project Osmos) by Naresh Venkat · 7 min read Introduces Project Osmos (preview), Fabric's data engineering agent: a purpose-built orchestrator that carries out persistent, collaborative data-engineering tasks inside Fabric, within the resources and boundaries you set, and validates its own output so teams can trust results rather than just code that runs.
  • The New Unified Copilot in Fabric and Project Osmos: From AI Assistance to Agentic Data Engineering by James Serra · 16 min read The first of a three-part series on AI in Fabric. James Serra explains the new unified Copilot in Fabric introduced at FabCon (a single AI companion across the data lifecycle) and how it differs from Project Osmos, the Fabric data engineering agent that carries out longer, multi-step data-engineering tasks within boundaries you set, with a medallion-architecture example.
  • Can You Trust Your Fabric Data Agent? Programmatically Testing Accuracy and Determinism by Marc Lelijveld · 9 min read This article explains how to programmatically test the accuracy and determinism of Fabric Data Agents using dynamic runtime calculations for expected answers to ensure reliable validation over time as configurations change.

📊 Reporting & Insights

  • Bring insights and action together with Fabric Apps in Org apps by Sunitha Muthukrishna · 3 min read How to include Fabric Apps in Power BI org apps, so users can go from an insight in a report to acting on it (submitting a request, assigning an owner, tracking a resolution) in a purpose-built Fabric App, all from one curated destination.
  • From a prompt to an operational dashboard: Generate Real-Time Dashboards with Copilot (Preview) by Michal Bar · 4 min read Copilot in Fabric can now generate a Real-Time Dashboard (preview) from a natural-language description of what you want to monitor, using the tables in a selected Eventhouse database; you then review and customise it. Walks through a stadium example comparing transit arrivals with gate activity.
  • Direct Lake on a Fabric Warehouse: How Loading Affects Performance and Cost by Vojtech Sima · 1 min read Vojtech Sima measures, on an F32 capacity, how loading a Fabric Warehouse with COPY INTO, CTAS or pipelines, plus V-Order and row order, shapes the Parquet files it writes, and what Direct Lake queries then cost.
  • All you need is a Fabric app by Mimoune Djouallah · 7 min read Mimoune Djouallah introduces OneLake Studio, a free VS Code extension he built to see how far a Fabric app can go with just the OneLake catalog: browse OneLake, inspect tables, open and explore semantic models, translate DAX to SQL and publish the result as a Fabric app, with DuckDB running in the browser.
  • Building Claude Dashboards on Power BI Semantic Models: Step-by-Step Setup, RLS, Sharing and Limitations by Soheil Bakhshi · 1 min read Soheil Bakhshi's step-by-step guide to building Claude Dashboards on top of existing Power BI semantic models, without opening Power BI Desktop or writing DAX, covering setup, row-level security, sharing and limitations.
  • Replaying El Clásico live in a Fabric App! by Nikola Ilic · 8 min read The project uses Microsoft Fabric, Rayfin SDK, and Python to recreate a historic Barcelona vs Real Madrid match in real time, processing StatsBomb events with KQL functions and displaying predictions through a React front end.
  • How to Use Direct Lake Across Public and Private Link Workspaces in Microsoft Fabric by Gilbert Quevauvilliers · 1 min read Gilbert Quevauvilliers shows how to use Direct Lake between a public workspace and a Private Link workspace in Fabric, a scenario Microsoft's documentation says isn't supported as a direct connection.
  • Build clearer, more expressive Real-Time Dashboards with new design and KPI capabilities (Generally Available) by Naor Biton · 6 min read Now generally available in Fabric Real-Time Dashboards: Design Tiles for titles, text and structure, multi-KPI tiles that group related measures from one query, and KPI styling that makes status, scale and targets easier to read.
  • On-premises data gateway October 2026 release by Goutham Upadhyaya · 4 min read The October 2026 release of the on-premises data gateway version 3000.338 improves accessibility with better keyboard navigation, clearer screen-reader announcements, and high-contrast support, preserves Power Query settings during tests, and offers detailed diagnostics for service-start failures.
  • DAX LLM Benchmark v2 by Marco Russo · 6 min read The updated DAX LLM Benchmark v2 reveals that recent frontier models excel at writing DAX, with model generation proving more influential than vendor distinctions, according to its creators.

⚙️ Data Engineering

  • Create an API in Azure by Laura Graham-Brown · 5 min read Laura Graham-Brown builds a simple, deliberately unsecured demo API with FastAPI and Python and deploys it to Azure Container Apps, so attendees at her Fabric workshop can pull sample product data with a copy job instead of uploading CSV files. For demo data only.
  • More ways to use scalar UDFs in Microsoft Fabric Data Warehouse (Preview) by Srdjan Matin · 4 min read Updates to scalar UDFs in Fabric Data Warehouse and the SQL analytics endpoint (preview): BREAK and CONTINUE in loops, large-value data types in more query shapes, UDF calls in IF EXISTS, existing functions picking up new capabilities automatically, and default inlineability checks with an explicit WITH INLINE = AUTO override.
  • GPU-Accelerated Data Warehousing for the AI Era (w/ Maraki Ketema & Craig Peeper) by Reid Havens · 1 min read Livestream (30 October) in which Reid Havens talks to Microsoft's Maraki Ketema and Craig Peeper about GPU-accelerated Fabric Data Warehouse: how it is built for high-concurrency analytics driven by BI, applications and AI agents, benchmarks they say show up to 7x faster performance than competitors, and the workloads that benefit most.
  • Fabric Notebook Toolkit: What Coding Agents Can Reach by Matthias Falland The Fabric notebook toolkit enables coding agents such as Claude Code and GitHub Copilot CLI to run and edit live notebooks with your login, promoting interactive development while requiring approval controls to prevent unintended data changes.
  • Metadata-Driven Pipeline Limits in Fabric, and When Not to Build One by Matthias Falland Video on when a metadata-driven pipeline in Fabric pays off: many similar sources through the same connector (say, 200 ERP tables), not a handful of tables or sources that each behave differently. It covers the limits to design around, including the Lookup activity's 5,000-row cap, ForEach and per-pipeline activity limits, and splitting work into parent and child pipelines.
  • Data Wrangler and Dates by Laura Graham-Brown · 4 min read Laura Graham-Brown's Data Wrangler demo in Fabric broke on a CSV with dates in DMY, MDY and YMD formats. She walks through what Data Wrangler can and can't convert in Pandas and Spark notebooks, and the code fixes for each, including to_timestamp/to_date from pyspark.sql.functions in Spark.

🏛️ Storage & Platform

  • What a Domain Does, and What It Does Not | Microsoft Fabric by Matthias Falland Video on what a Fabric domain does and doesn't do: per Microsoft Learn, assigning a domain doesn't change who can see or access items, which still comes from workspace roles and item permissions, and domain names are visible across the tenant. What you do get is an attribute, a OneLake catalog filter and delegated settings. A label, not a wall.
  • What Is a Platform Workspace? Metadata-Driven Workspaces in Fabric by Matthias Falland Video on the platform workspace pattern for managing dozens of Fabric workspaces as metadata: one row per workspace, with a platform workspace holding the metadata, provisioning and orchestration but no business data. Names are computed, an API sits behind every step, and teams get roles only on their own workspaces.
  • Fabric Cross-Workspace: Shortcuts, Invoke Pipeline, Variable Library by Matthias Falland Video on what crosses a Fabric workspace boundary when one platform workspace creates and runs many others: data crosses through OneLake shortcuts under the caller's identity, and runs cross through the Invoke Pipeline activity and a connection, but configuration (variable libraries, bindings, branches) and workspace identities stay home. So the platform runs everything, and each workspace keeps its own configuration.
  • Is Cosmos DB Mirroring Slow? Snapshot vs. Stream in Fabric by Matthias Falland Video on why Cosmos DB mirroring in Fabric can seem slow: there are two clocks, a bulk initial snapshot that scales with your data volume and a near-real-time stream. Unlike Synapse Link there is no analytical TTL, so deletes (TTL expiries included) land immediately: a mirror is a replica, not an archive.
  • Cosmos DB Mirroring Schema: How Fabric Handles Schemaless Data by Matthias Falland Video on how Cosmos DB mirroring in Fabric handles schemaless data: the mirrored table is a union of every property name seen, so a renamed property becomes two columns, missing values arrive as silent nulls and type mismatches fall to null. Treat the document shape as an interface.
  • Cosmos DB Mirroring: Why Nested JSON Arrives as a String by Matthias Falland Video on why Cosmos DB mirroring in Fabric delivers nested JSON as a single string column rather than flattening it: a schemaless source has no fixed shape to flatten to. It shows expanding it with OPENJSON, and the difference between CROSS APPLY and OUTER APPLY.
  • How your lakehouse design decisions affect Direct Lake performance by Nikola Ilic · 18 min read Nikola Ilic writes one 500-million-row dataset to seven Delta tables, each varying file count, row group size, V-Order or partitioning, and queries them through one Direct Lake on OneLake model. Warm queries all return in under half a second, but cold queries range from 0.8 to 14 seconds depending on how the table was written.
  • Bringing Rust-native Spark compatibility to Microsoft OneLake with Sail and LakeSail by Amit Chandra, Shehab Amin · 7 min read Sail, an open-source, Rust-native drop-in replacement for Apache Spark with built-in Delta Lake and Iceberg support, connects natively to OneLake: existing Spark Connect code can read and write Fabric lakehouse tables with no data copies or extra connectors. Also covers Arrow Python UDFs for AI enrichment and LakeSail Cloud for running Sail in production.
  • Access versioned lakeFS data from Microsoft Fabric with Microsoft OneLake shortcuts by Amit Chandra, Iddo Avneri · 4 min read Using OneLake S3-compatible shortcuts to reach lakeFS-managed data from Fabric: point a shortcut at a live branch for up-to-date analytics, or pin it to an immutable commit when you need to reproduce an earlier analysis, without separate data-copy pipelines.

🔐 Governance & Security

  • Item creation blocked due to item creation policies by Aleksi Partanen Short clipped from Aleksi Partanen's Fabric policies video: when an item creation policy blocks Teemu Multanen from creating a warehouse, Fabric shows a clear message naming the policy as the reason, rather than a generic error, and he encourages giving feedback on helpful error messages.
  • Contributors can make policies disappear completely by Aleksi Partanen Short clipped from Aleksi Partanen's Fabric policies video: Teemu Multanen shows that contributors can delete a policy set item from its workspace, and add or change rules on an active policy without being a capacity admin, so restrict access to the workspace where policies are managed.
  • Too Many Workspace Admins? Branch Workspace Admin Profile (Preview) by Aleksi Partanen Short clipped from the Fabric September 2026 feature summary: Teemu Multanen and Heini Ilmarinen on the branch workspace admin profile (preview), which lets developers branch out into new workspaces without their own rights to create workspaces or assign capacity, with the admins named in the profile added to every branch workspace.
  • Who Can Build Fabric Apps? Microsoft Fabric Policies (Preview) by Aleksi Partanen Short clipped from the Fabric September 2026 feature summary: Teemu Multanen and Oskari Heikkinen on Microsoft Fabric policies (preview). Capacity admins can control who creates which item types (Fabric Apps included) on a capacity, and tenant-level policies cover who can edit workspace network-security and customer-managed key settings and who can share data externally.
  • Microsoft Fabric: Who sees what when Copilot asks? #Shorts by Matthias Falland Short: when Copilot answers questions over Fabric data, it uses each user's own permissions, so row-level security on the semantic model still applies. Put security on the model, not just in the report. Clipped from Matthias Falland's video on Fabric IQ in Microsoft 365 Copilot.
  • Strengthening data security with workspace-level private connectivity for custom endpoints for Eventstream (Preview) by Spurthi Kommajosula · 7 min read Eventstream now supports workspace-level private links for custom endpoint sources (preview), so applications in restricted networks with no route to the public internet can send real-time data into Fabric privately.
  • Using Allow workspace settings editing policy in Fabric (Preview) by Yael Biss · 4 min read The new Allow workspace settings editing policy in Fabric (preview) lets organizations precisely control which users can modify sensitive settings like encryption and networking while keeping everyday workspace management within team roles.
  • Power BI Administration at Scale: PPCC26 Wrap-Up and Tenant Settings Demo Code by Heidi Hasting · 5 min read Heidi Hasting wraps up her PPCC26 session with Shabnam Watson on Power BI administration at scale and shares the promised demo code: a notebook, lakehouse and report, built with AI's help, that snapshots tenant settings through the admin API to keep a history of changes and show who each setting is actually enabled for.
  • Govern your Microsoft Fabric data estate from one place with the OneLake catalog and Govern Skills (Generally Available) by Lee Benjamin · 6 min read The Govern experience in the OneLake catalog is now generally available, centralising governance for tenants, domains, capacities, workspaces, tags and policies with estate-wide insights and a customisable governance report. Govern Skills, added to the Skills for Fabric collection, give AI agents the knowledge to help audit and fix governance issues. Both were announced at FabCon Europe.

💰 Management & Cost

  • Can't run all the items in a deployment plan? Use pipelines by Aleksi Partanen Short clipped from Aleksi Partanen's deployment plan video: only a small subset of item types can run as actions in a Fabric deployment plan (preview), so Aleksi's workaround is to put data pipelines in the plan. Teemu Multanen adds that with pipelines and notebooks you could technically do anything, even build your own deployment plans in a notebook.
  • Deploy the lakehouse, then populate it by Aleksi Partanen Short clipped from Aleksi Partanen's deployment plan video: deploying a lakehouse in Fabric carries over its definition but not its tables, so a shortcut to a table that doesn't exist yet has nothing to point at. Teemu Multanen's fix is a post-deploy action in the deployment plan (preview), a notebook or pipeline that populates the lakehouse.
  • Should the deployment plan be an item or not? by Aleksi Partanen Short clipped from Aleksi Partanen's deployment plan video: should a deployment plan be a Fabric item? In the preview it is, as are Fabric policies, and Teemu Multanen is happy that this means the plan follows the workspace and is version controlled.
  • Create Fabric Workspaces With the REST API and a Service Principal by Matthias Falland Video walking through one metadata-driven provisioning run: a notebook reads requested rows from a control table and, for each, calls the Fabric REST API as a service principal to create the workspace with its capacity and domain, assign group roles, provision a workspace identity, optionally connect Git, and write the ID back so reruns skip it.
  • Manage once, reuse everywhere: Introducing variable library support for Eventstream (Preview) by Spurthi Kommajosula · 6 min read Eventstream now supports variable library references (preview) for supported sources and destinations. They are preserved through Git integration and deployment pipelines, so one eventstream definition can move from dev to test to production with environment-specific connections.

📅 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