Fabric Weekly

Issue 16

27th October 2026

🎃 Welcome to issue #16 of Fabric Weekly, and happy Halloween!

There's a really clear theme this week: Fabric is getting a lot easier to work with through coding agents! There's also a big batch of real-time updates, and loads of great community content to dig into.

On the official side:

From the community:

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

  • Tuesday 27 October (tonight), online - Devon & Cornwall Power BI with Marjolein Opsteegh on working with AI to build dashboards people understand.
  • Thursday 29 October, Manchester - PBIMCR with Martin Brierley and Ian Pike on their FabCon Europe takeaways.
  • Wednesday 4 November, Birmingham - #PBIBRUM with Ashley Warren on faster semantic models and Juliana Smith on making Power BI reports accessible.
  • Wednesday 4 November, Newcastle - Newcastle Power BI and Fabric with Brijesh Tanna on data testing frameworks and notebooks.

🤖 Copilot, AI & Agents

  • Fabric Data Agent in Foundry: Move to Fabric IQ by Nov 6
    Video explaining that Microsoft Foundry agents calling a Fabric data agent as a custom tool must switch to the Fabric IQ tool (preview) before 6 November 2026, when the old integration stops being supported. Walks through the switch and flags early blockers: cross-geo processing and storage for AI must be enabled, and the workspace must be in a region with the full Fabric stack.
  • Microsoft Fabric Skills – What’s in it? When to use which skill?
    Wolfgang Strasser walks through the 25 skills in release v0.3.19 of Microsoft's open-source Skills for Fabric: reusable instructions that help AI coding assistants work with Fabric, grouped across Spark and lakehouse engineering, SQL, Dataflows, Real-Time Intelligence, Power BI and Fabric IQ, governance and ALM, migration and cost estimation. He also highlights the new Project Osmos skill for long-running agentic data engineering.
  • Microsoft Fabric Skills – When to use which skill?
    In the second part of his Skills for Fabric mini-series, Wolfgang Strasser shares a decision graphic for picking the right skill, followed by a table for each of the 25 skills giving a typical question you might ask your AI assistant and an example use case.

📊 Reporting & Insights

  • Power BI Apps: Workspace Apps, Org Apps and Fabric Apps
    Valerie Junk explains the difference between workspace apps, org apps and Fabric Apps, which all appear on the Power BI Apps page, and when you'd use each one.
  • Data Visualization in the Agentic Era
    Alper, from the Power BI Core Visuals team, introduces the Fabric visuals and Fabric data grid React component libraries (preview, on npm). Built on Vega-Lite and shipped with the Fabric Apps analytics template, they give developers and AI coding agents Power BI-style defaults, theming, data-grid features and a shared data and interaction model, so generated data apps don't have to rebuild these basics.
  • Teemu Vibe Coded a Power BI Report (Sort Of)
    In this Short, Teemu Multanen describes trying the data app template in Fabric Apps (preview): his coding agent connected it to a semantic model, though the result wasn't the Power BI experience. Oskari Heikkinen notes you can always go through a semantic model when your data source isn't directly supported.
  • Create purpose-built custom visuals in Real-Time Dashboards with Copilot (Preview)
    Copilot's preview feature enables users to create custom real-time dashboard visuals by simply describing their operational questions in natural language, which then generates tailored KQL queries and HTML visualizations for specific workflows or audiences.
  • Getting Started With Fabric Apps
    Taylor Amy and Teemu Multanen build and deploy a first Fabric app (preview) from Microsoft's to-do template, covering project structure, built-in sign-in, the SQL database and GraphQL API, secrets, capacity use and CI/CD. Fabric apps need a tenant admin to enable them, aren't in every region yet, and are announced but not yet available for Power BI Pro.
  • Install DAX libraries into Power BI Desktop by asking your AI assistant
    Marco Russo introduces version 0.1.0 of the DAX Lib MCP server, which lets an AI assistant search the DAX Lib catalogue and install user-defined function packages directly into a model open in Power BI Desktop, using the original code. It runs on Windows x64 only and needs compatibility level 1702 or higher.
  • Continuous axis for any Date table in Power BI
    Video companion to SQLBI's article from last week: Marco Russo shows how to get a continuous month-level X-axis in Power BI with any Date table, as the auto date/time feature does, by setting the right metadata on the Date table and its hierarchies.

⚙️ Data Engineering

  • Metadata-Driven Control Table in Fabric: What a Row Holds
    Video on what one row of a metadata-driven pipeline's control table holds (source object, target table, layer, full or incremental load, watermark column, active flag), plus the source-systems and run-log tables around it. The rows live in a SQL database in Fabric, a variable library holds the environment settings, and the rows are scripted into Test and Prod as code.
  • Fabric Incremental Load: Watermark and Log Table, Step by Step
    Video stepping through one incremental-load iteration of a metadata-driven pipeline in Fabric: read the old watermark, read the new one first, copy the range between them, write the watermark only after success, and log every iteration. Also points out that a watermark never sees deletes.
  • Materialized Lake Views for Files: Bronze in One Statement
    Video on file ingestion with Materialized Lake Views (preview): one SQL statement turns a folder of CSV or Parquet files in OneLake into a bronze table, with no pipeline or notebook in front of it. Covers when it fits, when it doesn't (database or API sources, Python logic, sub-second updates), and edge cases such as paths with spaces and nested folders under shortcuts.
  • What Is FMD? Erwin de Kreuk's Fabric Metadata-Driven Framework
    Video introducing FMD, Erwin de Kreuk's open-source (MIT) Fabric Metadata-Driven Framework. It loads source tables through a landing zone (files), Bronze (current state) and Silver (full history), driven by rows in one configuration database, and stops at Silver, so the Gold model is yours to build.
  • How FMD by Erwin de Kreuk Loads a Fabric Table: Landing to Silver
    Video following one table (Orders) through FMD, Erwin de Kreuk's Fabric Metadata-Driven Framework, from source to Silver: one orchestrator pipeline runs the three layers in order, views generate the queries, queues sit between the layers, and the watermark is stored only after the queue step.
  • FMD by Erwin de Kreuk: How to Add a New Source Table in Fabric
    Video on adding a new source table to FMD, Erwin de Kreuk's Fabric Metadata-Driven Framework: you add rows to the integration schema of its configuration database (a connection, a data source and three rows) rather than editing a pipeline, and the execution and logging schemas handle the rest, with credentials kept in Fabric.
  • FMD by Erwin de Kreuk vs Fabric Copy Job: Limits, When Not to Use It
    Video comparing FMD, Erwin de Kreuk's open-source Fabric Metadata-Driven Framework, with Fabric's Copy job: use Copy job first where its CDC and SCD Type 2 support covers your sources, and FMD when you want a table registry and history without CDC. Also notes that incremental loads can't see deletes and that you run FMD yourself.
  • Connect coding agents to Fabric notebooks with FNTK (Preview)
    The Fabric Notebook Toolkit (FNTK, preview) is a command-line tool that lets coding agents such as Claude Code, GitHub Copilot CLI and Codex find Fabric notebooks, read and edit cells, run one cell at a time in a live session, and pull back cell output, Spark job state and driver, executor and Livy logs for debugging. It installs from PyPI.
  • Announcing workspace monitoring integration for Fabric Data Warehouse (Preview)
    Fabric Data Warehouse now integrates with workspace monitoring (preview). Completed, user-initiated queries from warehouses, SQL analytics endpoints and warehouse snapshots land in a WarehouseExecutions table in the workspace's monitoring KQL database, so you can compare query durations, find recurring failures, and use OperationId to jump to the matching Query Insights record.
  • Simple Report Bursting With Fabric Dataflows Gen2
    Chris Webb combines the new Send Email action in Dataflows Gen2 with parameterised dataflows and a pipeline ForEach loop to send a personalised email to each recipient. He finds it slow, and it fails if the loop runs concurrently, so it only suits a handful of emails until HTML email bodies arrive (on the roadmap).
  • From mixed events to real-time action: Fabric Eventstreams Q3 2026
    Eventstreams' Q3 2026 roundup. Now generally available: SAP Datasphere, Cribl, Solace PubSub+ and MQTT connectors; Mirrored Database Change Feed and MongoDB CDC connectors; Schema Registry and Event Schema Sets; and Activator rule management. In preview: Custom Stream Connector (Kafka Connect), schema-aware Eventstreams, Reference Data joins with Delta tables, Workspace Monitoring, a Capacity Operation Events source, workspace-level Private Link, Variable Library support and Copy Job support.
  • Explore and process multiple event types with Schema-aware Fabric Eventstreams (Preview)
    Schema-aware Eventstreams (preview, new Eventstreams only) lets one stream carry several event types, whether they use registered schemas, are classified by a field such as deviceType, or are untyped. A redesigned data preview shows each event shape, and operators and destinations can then work on a single event type.
  • Incrementally migrate and re-migrate database objects with Migration Assistant for Fabric Warehouse (Preview)
    Migration Assistant for Fabric Data Warehouse now supports incremental migration (preview): move database objects in phases into an empty, previously migrated or existing Warehouse, resolving name clashes side by side. Sources are SQL Server, Azure Synapse dedicated SQL pools and Teradata. Only object definitions are migrated; data loading is a separate step, for example with a copy job.
  • Simplify Teradata migrations with Fabric Migration Assistant (Preview)
    Fabric Migration Assistant for Data Warehouse now supports Teradata sources (preview). It converts .sql and BTEQ files uploaded as a ZIP into Fabric-compatible T-SQL for tables, views, stored procedures, SQL functions and macros, lists anything that needs fixing, and Copilot can suggest corrections. Data is loaded separately, using a copy job, COPY INTO or pipelines.
  • More expressive graph analysis: GQL improvements in Microsoft Fabric Graph
    The January-August 2026 updates to GQL in Microsoft Fabric Graph introduce features like ANY SHORTEST for shortest paths, quantified edge patterns for conditional traversals, and unbounded variable-length paths, enabling more precise control over graph queries and analysis of complex connections.
  • T-SQL language enhancements in Microsoft Fabric Data Warehouse (Preview)
    Fabric Data Warehouse adds four T-SQL language enhancements in preview: GROUP BY ALL, ORDER BY ALL, the QUALIFY clause for filtering on window-function results, and a source-first FROM … SELECT query format.
  • New analytical functions in Microsoft Fabric Data Warehouse (Preview)
    Fabric Data Warehouse and the SQL analytics endpoint gain MEDIAN, QUANTILE, APPROX_MEDIAN and APPROX_QUANTILE in preview, usable as aggregate or window functions. The existing PERCENTILE_CONT and PERCENTILE_DISC functions now also support aggregate syntax.

🏛️ Storage & Platform

  • Read OneLake Tables From Your Third Party Tool, OneLake Security Included (Preview)
    A Short on the new OneLake table read API (public preview), which lets third-party tools read Delta Lake and Iceberg tables from OneLake with OneLake security's row- and column-level rules still applied.
  • Get Your House in Order: Microsoft Fabric Workspace Architecture
    Video: Alex Ward walks through organising a Fabric environment with domains, tags, workspaces split into ingest, transform and serve, and task flows, for easier governance and navigation in the OneLake catalog.
  • Fabric Mirroring for Azure Databricks: What Crosses, What Doesn't
    Full Fabric Friday episode (about 35 minutes) on connecting Azure Databricks and Fabric: mirroring, publishing, OneLake catalog federation and Spark over ABFS. Shows that mirrored data arrives but Unity Catalog grants don't, with demos of mirroring a catalog, Direct Lake over the mirrored data, and Databricks reading and writing OneLake, plus who pays.
  • Snowflake Iceberg Tables in OneLake: One Copy, Two Engines
    Video on OneLake and Snowflake interoperability (generally available since February 2026): Snowflake can store Iceberg tables directly in OneLake and read Fabric tables as Iceberg. Limits include public-network-only access (no private link or on-premises gateway), matching regions for writes, and trouble with tables that take frequent small commits.

🔐 Governance & Security

  • The Govern Tab Shows Tenant Settings Now… Unless You’re Not an Admin
    Heidi Hasting finds the OneLake catalog's Govern tab now shows Tenant settings and other platform configuration, but only to tenant admins; non-admins see just Known issues, Service health and Support requests. She argues for a read-only view of tenant settings for all users, and suggests extracting them via the admin APIs and sharing them in the meantime.
  • Half a year later, there's notebook one, two, three in production
    In this Short, Teemu Multanen explains how Microsoft Fabric policies (preview) can limit item creation in production workspaces, so ad-hoc notebooks created while investigating errors don't build up over time.
  • Create one allow rule and anything not allowed is blocked
    In this Short, Teemu Multanen explains a gotcha in Microsoft Fabric policies (preview): adding your first allow rule turns the policy into an allow list, so every operation that doesn't match an allow rule is blocked.
  • #9 Planning a Fabric Solution Fit for Your Organization
    Part 9 of Paul Turley's Getting Started & Tenant Foundations series argues for adopting only the Fabric workloads that solve a real business problem (most organisations start with Data Factory and Power BI), sizing capacity to data volumes, concurrency and workloads rather than headcount by measuring peak use on a trial capacity in the Capacity Metrics app, and rolling out in phases while leaving room to split workloads across capacities later.

💰 Management & Cost

  • Measured, not estimated: what Real-Time Intelligence in Microsoft Fabric really costs
    Rene Fürstenberg measured a complete Real-Time Intelligence solution in Fabric for 24 hours: 76 CU hours a day, too much for an F2 (which hit interactive rejection on day two) but about 80% of an F4. Eventstreams bill per route rather than per transformation (two routes were nearly half the load), a continuously fed Eventhouse never suspends, Activator charges per active rule, dashboard viewing added nothing measurable, and each data agent question cost about 1,400 CU seconds.
  • Using Power Automate to Switch a Workspace Capacity
    Laura Graham-Brown builds a Power Automate child flow that calls the Fabric REST API's Assign to Capacity endpoint, using HTTP with Microsoft Entra ID, to move a workspace between two capacities in the same region, for example ahead of a peak or when a capacity is running hot.
  • Lift and shift vs modernizing your data warehouse
    James Serra weighs lift-and-shift, replatforming and modernisation when migrating a data warehouse to Fabric, for example from Netezza or a Synapse dedicated SQL pool. Even a conservative migration takes real discovery and validation work; his preference is selective, phased modernisation, with any 'phase two' given an owner and funding up front.
  • Super complex deployment plan? Is your architecture a bit weird?
    A Short on Microsoft Fabric deployment plans (preview): if your deployment plan needs many steps, it may be worth questioning your architecture. One handy use case is deploying configuration for metadata-driven frameworks.
  • What if you just select the deployment plan?
    In this Short, Teemu Multanen explains a gotcha with Microsoft Fabric deployment plans (preview): selecting the plan alone doesn't select the items it references, so you also need to select the items you want deployed.
  • F0 for Your Playground? Just Don't Throw In Eventstreams
    A Short on F0, a coming-soon zero-provisioned Fabric capacity with on-demand billing. Teemu Multanen and Heini Ilmarinen warn that always-on items such as Eventstreams can run up costs in a test environment.
  • Ep. 31 - Piloting Fabric at a Nonprofit
    In episode 31 of the Figuring Out Fabric podcast, Eugene Meidinger talks to Olena Kulish, who runs digital analytics at the National Council on Aging, about piloting Fabric at a nonprofit where every tooling dollar has to be justified to donors: why she ran a pilot rather than going all in, how hard it is to estimate capacity during a 60-day trial, and having to rebuild everything when the region they needed wasn't available. They also cover the DP-600 exam and why AI readiness is mostly the basics.
  • Fabric Deployment Plan - Ep.573
    Podcast episode: Mike Carlo and Tommy Puglia discuss Fabric deployment plans, which let you order deployments and run notebooks or pipelines before and after each item, which teams should use a plan instead of Git and GitHub Actions, and why both would wait for general availability.

📅 Community & Events