Fabric Weekly
Issue 8
25th August 2026
👋 Welcome to issue #8 of Fabric Weekly!
This week, on the official side:
- Matt Hicks announces Mirrored Google Lakehouse runtime catalog, now in preview - bring Google-cataloged Apache Iceberg tables straight into OneLake, zero-copy, joining the existing AWS Glue and Azure Monitor mirroring sources.
- Sumiran Tandon on Customer-Managed Key APIs for Fabric workspaces, now generally available - assign, inspect, rotate and reset workspace encryption entirely through REST, with a tenant-wide admin view for governance at scale.
- Salil Kanade on bringing agentic warehouse development to Fabric Data Warehouse (preview) - a remote MCP server for executing T-SQL plus Skills for Fabric, so an agent can move from drafting a query to validating it against live data without treating each step as a separate tool.
- Spurthi Kommajosula on workspace identity for Azure Event Hubs sources in Eventstream (preview) - authenticate via the Fabric workspace's own identity instead of a shared access key, so credentials never sit in your Eventstream config.
From the community:
- Marco Russo (SQLBI) on testing DAX measures by using AI - connecting an AI assistant to the Power BI Modeling MCP server to discover positive, negative, and false-positive/negative test cases for a dynamic measure, then wiring them into DAX user-defined functions for repeatable regression testing.
- Chris Webb rounds up what's new in Semantic Link Labs' interactive tooling - lineage exploration and a Vertipaq Analyzer view, both usable from a Fabric notebook with barely any code.
- Jake Duddy shares musings on agentic development in Fabric - lessons from building an agentic RSS-recommendation project end to end, on why Fabric-specific skills matter more than spec-driven frameworks, and why the Fabric CLI beats fabric-cicd for the early, iterative agentic loop.
- Meagan Longoria lists ten structural gaps in Fabric deployment pipelines, with alternatives - from deployment rules you can't set until after a first deploy, to no approvals or post-deployment hooks, plus three ways to work around it.
- James Broome (endjin) on DAX calculated tables for row-level security context - pre-aggregating comparative benchmarks into a separate table so restricted users can see how they're doing against a peer group without RLS ever exposing anyone else's raw rows.
And, some upcoming events:
- Tuesday 22 September, online - Devon & Cornwall Power BI on agentic development in business intelligence, a beginner's guide from Kurt Buhler covering the building blocks of AI and agents, and where self-service BI is headed.
- 28 September - 1 October, Barcelona - FabCon + SQLCon Europe 2026 lands at the CCIB, and several of the UK groups are heading out there.
- Reminder - Thursday 27 August, Salford - PBIMCR's Ghost in the Aether, a murder mystery built entirely on Microsoft Fabric, from David Mitchell and Adrian Critchley.
- Reminder - Wednesday 2 September, Birmingham - #PBIBRUM host Vlad Mihanta on Power BI Embedded, and Neil Rowlands on building a Power BI training programme.
🤖 Copilot, AI & Agents
- Musings on Agentic Development in Fabric Reflections from building an agentic RSS-recommendation project on Fabric (Spark notebooks into a Lakehouse, LLM classification and embeddings, k-means clustering, a SQL Database, and personalized recommendations via a Power BI translytical task flow). Covers coding-agent fundamentals (harnesses, context rot, Copilot customization via instructions/prompts/skills/agents), argues skills are essential because LLMs don't know Fabric item structures, and concludes spec-kit felt redundant for this project while fabric-cicd's vague deployment errors argue for using the Fabric CLI during early development and saving fabric-cicd for once the solution is stable.
- Tools With Interactive UIs In Fabric Notebooks With Semantic Link Labs Semantic Link Labs enhanced its Fabric notebook tools with an interactive UI for visualizing report and semantic model lineage with minimal coding, featuring Vertipaq Analyzer integration.
- AI Agents Now Do My Fabric Data Engineering The video shows how Claude handles development tasks while Codex reviews code in a seamless data engineering workflow on Microsoft Fabric, demonstrating secure service identities and cross-vendor collaboration.
- Enhanced data agent visualizations with Fabric visuals (Preview) Fabric Visuals improves data agent queries with better charts, consistent styling, and faster insights.
📊 Reporting & Insights
- Modern DAX in Power BI: UDFs, Window Functions, INFO & More (Part 4) Part 4 of Reid Havens’ DAX series covers reusable user-defined functions, advanced window functions, INFO functions for metadata querying, and debugging tools while noting their capabilities and limitations.
- Testing DAX measures by using AI The article explains how AI automates identifying relevant test cases for DAX measures, ensuring business logic remains correct after changes.
- Building to Teach: AI-Assisted Tools for Data Viz Education (w/ Ben Ferry) Ben Ferry shows how AI tools like Claude Code let non-engineers create data visualization education resources, exemplified by Ferry Tales Studio, a browser-based Vega-Lite chart editor for Power BI and Deneb users.
- Power BI Legend Sorting: Sort by Value While Keeping a Series in Place (Taglish) Learn how to dynamically sort Power BI legend items by measure value while fixing specific series positions using a DAX workaround.
- SVG Rendering Cost in Power BI Testing SVG rendering cost in Power BI (prompted by a DAXLib session and Chris Webb's Performance Analyzer talk) found most element types (filters, gradients, animation) render similarly, but rendering many elements compounds cost — 10 tables of 10,000 circles each took 5 seconds to render, 7 seconds when combined into one table. Practical advice: optimise the DAX, reuse elements with defs/use, avoid animations and complex filters, and limit SVG visuals per page.
- DAX Performance Tuner MCP Server Using DAXNoob's DAX Performance Tuner MCP Server to optimise the expensive DaxLib.SVG.Compound.Violin() function via Claude Sonnet 4.5, GPT-5.2, and Gemini 3 Flash. The Gemini variant was dropped after its violin plots rendered incorrectly, leaving Claude and GPT-5.2 for DAX Studio benchmarking: the original took 1907ms (FE 1700ms, SE 207ms) versus 923ms for the GPT-5.2 variant (FE 860ms, SE 63ms) — a 48% improvement, achieved mainly by dropping an unneeded axis-index table scan. Takeaway: useful for spotting oversights, but still needs rigorous testing and human validation.
- Build a Cohort Analysis Dashboard in Power BI | Step-by-Step This guide walks through creating a Cohort Analysis Dashboard in Power BI, detailing the logic for tracking customer retention and churn over time using DAX.
- Monitor key metrics at a glance with the new KPI visualization in Real-Time Dashboards (Generally Available) The new KPI visualization in Microsoft Fabric's Real-Time Dashboards adds scale, status ranges, formatting options, and reference indicators to quickly convey metric health at a glance.
- Building Custom Data Experiences with Power BI and Fabric Apps (w/ Stephen Godderidge) The livestream with Stephen Godderidge explains how Fabric Apps extend Power BI by offering customizable, interactive data experiences built directly on top of Fabric’s semantic models.
- Providing comparative context with DAX Calculated Tables under Row Level Security This article explains how DAX calculated tables provide comparative context under row level security by pre-aggregating data for restricted users to view meaningful benchmarks without exposing sensitive information.
⚙️ Data Engineering
- Stop Duplicating Pipelines! Use Schedule Parameters in Microsoft Fabric Learn how to eliminate pipeline duplication in Microsoft Fabric by using schedule parameters, allowing a single pipeline to dynamically load weather data into both UK and USA tables with minimal code.
- I Gave AI My Fabric Admin Account. Do Not Copy This Part A live admission from a Fabric build: the AI agent authenticates as the user, so it inherits everything their account can see. The real protection isn't a guardrails file — it's thinking carefully about what access and permissions you actually give your agent.
- It Followed My Naming Conventions. I Never Told It To The video shows how a markdown file with naming conventions auto-applies these standards to AI-generated data pipelines in Microsoft Fabric, ensuring consistency without manual prompts.
- The AI Broke My Demo Live. Then It Fixed Its Own Bug During a live demo of Microsoft Fabric, an AI agent unexpectedly identified and fixed a bug in the notebook by dynamically resolving the workspace name at runtime.
- They Called the CLI Last-Century Throwback Stuff. Then She Looked Closer Ginger Grant shows CLI's power in automation and integration for managing multiple MCP servers despite common beliefs that it is outdated compared to UI interfaces.
- She Develops for Fabric with the Capacity Turned Off Ginger Grant shows a feasible Fabric development workflow using only Claude for local coding help and manual GitHub pushes, without Fabric Compute Units or GitHub Copilot.
- Fabric Dev Setup: Java, Miniconda, PATH... or Just Connect Your Repo Configuring VS Code for Fabric engineering the traditional way means installing Java and Miniconda and wrangling your PATH. Ginger Grant's Claude-based setup was just "connect your repo" — no extra installs required.
- Microsoft Stopped Building CDC in ADF. It Became Fabric's Copy Job Azure Data Factory's change data capture did not go to GA as development stopped; the feature moved to Fabric evolving into the Copy job, making Fabric an enhanced version of ADF.
- Nobody Writes Code Anymore. Microsoft's PM Says That's Exactly Right Mark from Microsoft states that nobody writes code anymore, highlighting the rise of AI-assisted vibe coding where users outline goals and tools produce notebooks, pipelines, and tests automatically.
- Bringing agentic warehouse development to Fabric Data Warehouse with MCP and Agent Skills (Preview) Fabric Data Warehouse launches MCP Server (remote) and Skills for Fabric, allowing AI agents to run SQL queries, verify results, diagnose problems, and utilize Fabric-specific insights without needing extra security setups or local servers.
- Building the Bronze → Silver → Gold layers Building the Bronze Silver Gold layers in Fabric Data Warehouse defines each layer’s job to ingest raw data apply cleaning and validation, and deliver business-ready insights using T-SQL patterns staging schemas and scaling practices.
- Connectivity patterns in Microsoft Fabric: A guide for data integration workloads This guide explains how to choose between VNet data gateways, on-premises data gateways, or direct cloud connections for connectivity in Microsoft Fabric based on source location and the need to cross private network boundaries.
- FabCon + SQLCon 2027: Selecting the right track for your session submission The call for content for FabCon + SQLCon 2027 (Atlanta, March 8-12) is open, and this post defines all ~18 submission tracks — from Power BI, OneLake, and Fabric Data Warehousing to Real-Time Intelligence and Fabric Apps — to help prospective speakers pick the track that best matches their session.
🏛️ Storage & Platform
- Bring your Google Lakehouse Runtime Catalog data to OneLake (Preview) The new Mirrored Google Lakehouse runtime catalog item enables Fabric users to access Google-managed Apache Iceberg tables directly in OneLake without data duplication, streamlining multi-cloud analytics.
- Microsoft's Pipeline PM: Don't Make Pipelines Your Default Mark advises against using pipelines as a default solution recommending Mirroring for free data replication and cost elimination.
- You Can Hit Fabric's 1,000-Item Limit Quite Easily The video explains how quickly teams can encounter Fabric’s 1,000-item workspace limit when using source-specific lakehouses for many systems and associated notebooks and pipelines.
- No SELECT * in My Data Loads. I Define the Schema. The video argues against using SELECT * in data loads, advocating for explicit column selection to define the schema more accurately and efficiently.
🔐 Governance & Security
- Dynamic Data Masking in Power BI Warns against the common DAX-measure pattern for masking sensitive fields in Power BI Import models — the underlying column is still queryable via Build permission, XMLA, or Analyze in Excel, so it isn't real security. Instead, split sensitive fields into a separate "satellite" table secured by RLS, with masked and unmasked duplicate rows per record (flagged by an isMasked column), so the security role — not a report visual — decides which version a user sees. Trade-off: a larger table and careful design to avoid grouping/totals issues.
- Secure Azure Event Hubs Connections in Eventstream with Workspace Identity (Preview) Workspace identity in Eventstream enables Azure Event Hubs connections without shared access keys, simplifying management through Fabric workspace identities and Azure RBAC.
- Customer-Managed Key (CMK) APIs for Microsoft Fabric Workspaces (Generally Available) Microsoft Fabric provides GA CMK REST APIs for managing encryption keys in Azure Key Vault programmatically, allowing assignment, inspection, rotation, resetting, and auditing of key usage across workspaces at scale.
- How to Master Microsoft Fabric Auditing and DLP Mastering Microsoft Fabric’s auditing and data loss prevention (DLP) capabilities ensures real-time compliance monitoring as users actively engage with enterprise data.
💰 Management & Cost
- Using Custom SQL Pools to balance performance and cost in Fabric Data Warehouse Custom SQL Pools (preview, soon GA) let admins cap the share of warehouse resources a workload can use — trading longer runtimes for a lower peak allocation and fewer billed vNode-seconds under allocation-based billing. Explicitly not a spending cap or budget limit, and best suited to non-latency-sensitive workloads like ETL and background processing.
- CICD for Fabric Data Teams A tour of Fabric CI/CD options from low-code to full API control: Deployment Pipelines alone lack branching, review, and fine-grained parameterization; adding Git Integration brings version control and PR review; dropping Deployment Pipelines for pure branch-based promotion adds environment parity and git-revert rollback; and for complex solutions, API tooling (Terraform, fabric-cicd, Fabric CLI) with a solution-per-repo structure gives scoped ownership, blast-radius containment, and agent-friendly repos. Recommendation: start with git integration for the cheap wins, add sophistication as complexity grows.
- Fabric-CICD Updates - Semantic Models, Parameters and Config Covers fabric-cicd v0.1.17 through v0.1.34: configuration-based deployment via config.yml, parameter file templates with an extend key for reuse across workspaces, dynamic parameterization (including cross-workspace item references), key_value_replace for precise JSON key replacements, and Semantic Model Binding — mapping semantic models to specific data connections, including on-premises gateways and multiple connections per model.
- Fabric-CICD - It's Official fabric-cicd v0.2.0 marks the tool's transition to official Microsoft support, with a committed long-term roadmap and deeper integration into the Fabric platform.
- Fabric App Documentation Site A walkthrough building a documentation site with the Zensical static site generator, served as a Fabric App via the Rayfin CLI (contributed as an awesome-rayfin zensical-docs template): enable Fabric Apps in the tenant, scaffold the project, deploy with rayfin up, wire the app URL into both zensical.toml and rayfin.yml for Fabric SSO, then iterate with a live-reload dev server.
- 10 Things I Hate About Fabric Deployment Pipelines— And Some Alternatives A detailed rundown of ten structural gaps in Fabric deployment pipelines — from tangled permissions and rules that need a first deployment before they work, to variable-library inconsistencies, broken item pairing, and no approvals or post-deployment hooks — plus three alternatives: scripting against the Fabric API, wrapping the pipeline in another CI/CD tool, or skipping pipelines entirely with fabric-cicd.
- When Your Microsoft Fabric Capacity Runs Out: Optimize, Scale Up, Scale Out, or Isolate? A framework for responding when a Microsoft Fabric capacity runs out of headroom: optimize wasteful CU usage first, scale up when healthy workloads collectively need more compute, scale out to separate workload classes onto different capacities, or isolate new/risky/mission-critical workloads with tryout, timeout, and rescue capacities — plus alerting and guardrails to catch pressure before users do.
- Billing personas for PowerTable within Planning in Fabric IQ (Generally Available) Fabric Plan now supports billing personas for PowerTable and Intelligence data management, enabling provisioning without Planning sheets, with editors billed at the Stakeholder tier and Viewers as read-only consumers.
- [SimplePBI] Obtener schedules-corridas de Fabric pipelines y notebooks [Spanish] This article shows how to use the SimplePBI library to retrieve schedule and run instance data for Fabric Data Pipelines and Notebooks via the Power BI REST API, allowing analysts to create dashboards and analyze workflows.
📅 Community & Events
- Devon & Cornwall Power BI User Group: Agentic Development in Business Intelligence — Tuesday 22 September, online Tuesday 22 September, online — Devon & Cornwall Power BI User Group: a beginner's guide to agentic development in business intelligence.
- FabCon + SQLCon Europe 2026 — 28 September–1 October, Barcelona 28 September – 1 October, Barcelona — FabCon + SQLCon Europe 2026; several UK user groups (including PBIMCR and Devon & Cornwall) are organising meetups on the ground.