Fabric Weekly

Issue 11

22nd September 2026

🍂 Welcome to issue #11 of Fabric Weekly!

This week, on the official side:

From the community:

And, some events in the next couple of weeks:

  • Tonight, Tuesday 22 September, online - Devon & Cornwall Power BI - Kurt Buhler on agentic development in business intelligence. It won't be recorded, so catch it live.
  • Thursday 24 September, Manchester - PBIMCR - Leon Gordon on monitoring and governing Fabric, and Vito Schittone on advancing your career in data.
  • 28 September - 1 October, Barcelona - heading to FabCon Europe? PBIMCR are hosting networking drinks on Wednesday 30 September.

🤖 Copilot, AI & Agents

  • Make Your AI Agent Less Creative. Seriously. by Aleksi Partanen The video shows how deterministic scripts can save tokens and manage AI agent creativity, improving efficiency in agentic CI/CD workflows.
  • What Does It Mean to Make Data AI-Ready? by James Serra · 7 min read Part 1 of James Serra's three-part series on AI-ready data, defining it as data that is accurate, well-described, semantically clear, governed, accessible and carrying enough business context for AI to interpret it correctly. Covers data quality and MDM, expanding acronyms and agreeing business definitions, metadata and relationships, governance, and the choice between changing the data and explaining it — with Fabric Data Agents and Power BI AI instructions as the places that context gets expressed.
  • Fabric IQ Confuses Everyone. Turns Out There's a Whole IQ Family by Aleksi Partanen Aleksi Partanen asks Roey Zalta to untangle Fabric IQ. It is one member of what Zalta calls Microsoft's "IQ family" — alongside Work IQ, Foundry IQ and the newly joined Web IQ — each covering a different part of Microsoft's intelligence layer. Fabric IQ is the one scoped to the Fabric environment, where your data sits alongside items such as ontology, semantic models, data agents and graphs.

📊 Reporting & Insights

  • Build Microsoft Fabric Data Apps with Rayfin CLI | Tutorial (2026) #microsoftfabric #Rayfin by Gaston Cruz, Alex Rostan Gastón Cruz and Alex Rostan (The Power Mates) build a Microsoft Fabric data app from scratch with the Rayfin CLI, Microsoft's visualization-as-code framework for data apps inside Fabric. They cover scaffolding with @microsoft/rayfin, writing DAX queries and Vega-Lite visual specs, deploying with `rayfin up`, and connecting to semantic models with full RLS support — all without touching the Fabric portal UI.
  • Find Unused Objects In Power BI Semantic Models With Semantic Link Labs by Chris Webb · 1 min read Semantic Link Labs introduces a feature that automatically identifies unused tables, columns, and measures in Power BI semantic models by analyzing downstream reports and Workspace Monitoring DAX queries, streamlining model optimization.
  • Power BI: the tale of migrating a report between two tenants by Reitse Eskens · 3 min read Power BI users encounter connection issues with migrating reports across tenants due to different Lakehouse GUIDs, though using exported PBIP files, edited TMDL scripts, and updated connection strings provides a workaround.
  • DAX UDFs, Arrow REST API, and Outbound Access Protection: A Conversation with Kay Unkroth (Fabric Insider Ep. 13) by Reza Rad · 13 min read Kay Unkroth describes how Power BI's new DAX UDFs enhance readability and reusability, the Arrow REST API boosts scalable DAX query execution, and Outbound Access Protection improves enterprise governance.
  • Moving off ODBC — A self-serve scanner for your ADBC migration by Michaela Isaacs · 4 min read Power Query is moving from legacy embedded ODBC drivers to Apache Arrow Database Connectivity (ADBC) across Power BI Desktop and the Fabric service, in two events per connector: a default flip that opens your validation window, then a cutover that disables ODBC entirely. To speed up the inventory step, Microsoft has published pq-adbc-advisor, a notebook-based scanner that finds semantic models, dataflows and Data Pipelines still pinned to the legacy driver and classifies each as Will fail, Needs review or Ready. Per-connector dates live in the migration guide rather than this post.

⚙️ Data Engineering

  • Nobody on This Panel Uses Fabric's Copy Job. Not Even Me by Aleksi Partanen A panel poll on Fabric's Copy job: Heini and Aleksi Partanen still build everything with the regular pipeline copy activity, and Oskari has only used Copy job on a hobby project. Teemu asks the better question — what signal would actually get teams to adopt it?
  • The Dataflow Gen1 Migration Wizard Comes With a Catch by Aleksi Partanen Fabric's new migration wizard may be the easiest way to move Dataflow Gen1 to Gen2, but Teemu warns that the original Gen1 dataflow is gone once migrated — and the wizard is still in preview, so think twice before running it on something you can't afford to lose. Gen1 is marked legacy with no new investment, but at recording time there was no published retirement date.
  • Upcoming changes to legacy Teams and Outlook activities in Fabric Data Factory pipelines by Connie Xu · 1 min read From 30 October 2026, legacy Teams and legacy Outlook activities can no longer be added to Fabric Data Factory pipelines, and will stop appearing in the add-activity list. Existing instances keep working and can still be edited in existing pipelines; new notification scenarios should use the modern Microsoft Teams and Outlook 365 Email activities.
  • A Runtime Upgrade Silently Broke My Data. No Errors, Just Nulls by Aleksi Partanen After a Spark runtime upgrade in Microsoft Fabric, an array-parsing function quietly changed behaviour: the notebook did not fail, but several columns came back as null. Aleksi Partanen's warning is that runtime upgrades change a lot under the hood, so test thoroughly before turning one on.

🏛️ Storage & Platform

  • Mirroring for Google BigQuery in Microsoft Fabric (Generally Available) by Michaela Isaacs · 1 min read Mirroring for Google BigQuery in Microsoft Fabric provides organizations a simplified, cost-effective method to integrate BigQuery data into their analytics ecosystem with production support and enterprise SLAs.
  • Why Microsoft Fabric beats Databricks (in the Microsoft world) by Aleksi Partanen Aleksi Partanen's case for Microsoft Fabric over Databricks inside a Microsoft shop: Fabric is more approachable for people sitting closer to the business, and it is ready as soon as you turn it on, where Databricks needs supporting Azure infrastructure you then manage yourself. He is explicit that Databricks is a strong platform built for engineers — the argument is about fit within the Microsoft ecosystem, not outright superiority.
  • Files Or Tables For Your Pipeline Config? by Aleksi Partanen The video explains a hybrid method using JSON files as the main configuration source in a lakehouse setup while also persisting them in tables for straightforward querying and metadata management.

🔐 Governance & Security

  • Customer managed keys can make you LESS secure by Aleksi Partanen Customer managed keys can be risky if regular rotation and strong governance are not maintained, possibly reducing security compared to using Microsoft’s managed encryption.
  • Soft delete for On-premises data gateways in Microsoft Fabric (Preview) by Goutham Upadhyaya · 3 min read Soft delete retains on-premises data gateway deletions for 30 days, enabling authorized admins to recover them before permanent removal in Microsoft Fabric (Preview).
  • OneLake Security: Setting Up Row Level Security From Lakehouse to Report by Reza Rad · 8 min read OneLake security enables defining row level security centrally so all related objects automatically inherit the configuration, simplifying governance across Fabric.
  • Fabric Core MCP Server: Scan and Configure OneLake Security with AI by Reza Rad · 6 min read Reza Rad walks through the Fabric Core MCP server, which exposes Fabric's APIs in a form an AI assistant can operate directly. Using GitHub Copilot in VS Code, he scans workspaces to find which lakehouses have custom OneLake security roles, reads back a role's row-level filter with the Fabric capacity paused, then has the AI create one role per country and reassign a test user's memberships — the kind of repetitive configuration work that would be tedious to do by hand.

💰 Management & Cost

  • The Hidden Cost of the Data Swamp: Fabric Capacity Guardrails by Pragati Jain · 3 min read Pragati Jain on how Fabric's pooled Capacity Units can quietly turn into a costly "data swamp": noisy-neighbour queries and notebooks throttling shared capacity, orphaned pipelines burning CUs over weekends, and unmonitored workspace sprawl. Her guardrails: isolate workloads on separate capacities, review the Capacity Metrics app weekly, alert before sustained utilisation reaches throttling thresholds, and apply lifecycle policies to abandoned workspaces and models.
  • I Asked 3 Fabric Experts If They Use Workspace Monitoring. Nobody Does by Aleksi Partanen A quick panel poll on Fabric's workspace monitoring, still in preview and billed on the capacity its monitoring Eventhouse consumes. Heini found it doesn't scale — switching it on everywhere would eat the whole capacity — and Oskari doesn't enable it by default. Only Teemu turns it on whenever a customer can afford it: great visibility, at a price.
  • #4 Fabric Licensing Explained: Per-User, Capacity-Based, and Trial Options by Paul Turley · 1 min read Part 4 of Paul Turley's Getting Started & Tenant Foundations series. Fabric licensing comes down to two separate decisions: who can create and share content, and what capacity powers it. Pro covers individual creators; Premium Per User adds features such as more frequent refreshes and XMLA endpoints but provisions no Fabric capacity, so lakehouses and notebooks still need an F-SKU. From F64 up, free users can view Power BI reports — the point where capacity tends to beat buying seats, which he puts at roughly 250–350 report users. The 60-day F64-equivalent trial is the cheapest way to test, and he advises choosing its region carefully, since moving Fabric data between regions later is painful.
  • I Turned Off Every Approval Prompt For My AI Agent by Aleksi Partanen The video shows running an autonomous AI developer agent with all approval prompts disabled, using bypass permission mode in a headless session while enforcing strict restrictions such as no main branch pushes or environment variable access via branch policies and service principals.
  • I Set A Trap For My AI Code Reviewer by Aleksi Partanen The video shows an AI code reviewer flagging an issue in a pull request while the developer agent is unaware of new pipeline concurrency rules, demonstrating effective inter-agent communication in Azure DevOps with Microsoft Fabric.
  • The query finished…Why is my Fabric SQL database still consuming CUs? by Nikola Ilic · 7 min read Fabric SQL database compute bills on the higher of CPU usage and a memory equivalent — 2.611 CUs per vCore, with memory converted at 3 GB per vCore — so a memory-heavy workload can show no CU saving at all after CPU is halved. A database holds a 2 GB memory floor while online and compute is only released after 15 minutes of inactivity, which is why Nikola Ilic's worked example has a two-minute application session followed by an idle period that consumes more than the session itself.
  • SQL Database for Fabric: Cost, Performance and Use Cases by Paul Turley · 1 min read Paul Turley reports back after using SQL Database in Fabric on several projects. It isn't billed per core or per database: it draws on your capacity's shared CU pool (1 vCore = 2.611 CUs), so it adds no new line item. But it isn't free either. A database stays online for about 15 minutes after the last query, holding at least 2 GB of memory, so a two-minute burst can bill around 17 minutes of compute. He maps each F-SKU to its nearest Azure SQL General Purpose size and recommends setting a max-vCore cap. The sweet spot is small-to-midsize operational stores: ETL metadata and logging, Translytical Task Flows and app backends. Anything heading past the 4 TB ceiling, or large but rarely queried, is a better fit for Azure SQL or a VM.

📅 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