Power BI Weekly

Issue 383

13^th November 2026

Just the one official announcement this week, but probably not a welcome one for some: Zoe Douglas has announced the retirement of R and Python visuals in Power BI in May 2027, after which they'll stop working in the service. Python scripts in Power Query are unaffected, for those of you whose data prep just flashed before your eyes!

Over in the community, Dániel Gábor Patkós has explained why your small model can fail to refresh on Fabric even when it refreshes happily on Pro, with tests showing that heavy Power Query work, not model size, drives peak refresh memory (so throwing a bigger SKU at the problem isn't always the fix). Elsewhere, Ilya Rzhevskiy has covered Databricks' new Publish to Power BI feature, which pushes Unity Catalog metric views into Power BI semantic models as DAX measures (one way only, so the definitions stay owned on the Databricks side).

Finally, Jiří Neoral has written a clear comparison of org apps and workspace apps, covering the automatic permission handling that makes org apps attractive, but the lack of versioning which means any changes in production go live the moment you hit "Save".

Data Modeling

Data agent ve Fabricu: kdy už Copilot nestačí [Czech] by Jiří Neoral · 7 min read When a Fabric data agent is worth it over Copilot in Power BI (answers that need several sources, or an agent called from Copilot Studio, Foundry or MCP), why calculation rules belong in Prep data for AI on the semantic model rather than in agent instructions, and hands-on test results: 6 of 15 combined model-plus-document questions answered correctly, at roughly 1,900 CU-seconds per question on an F2.
Your CFO Will Want This: Planning in Fabric by Bas Dohmen Bas Dohmen tours Planning in Fabric: building budgets, forecasts and scenarios in an Excel-like sheet with approval flows, comparing plan against actuals on the same semantic model, and writing results back to a SQL database in Fabric.
Using AI to Build Semantic Models, Ontologies, and Fabric Data Agents by James Serra · 11 min read The final part of James Serra's series on AI in Fabric uses a net-revenue example to show where Copilot in web modeling and the Semantic Model Authoring skill, the Ontology Agent and 'Build agent with AI' each fit, recommends starting with a data agent over a tested semantic model before adding an ontology, and flags preview status and a mismatch between Microsoft's setup docs and FabCon announcement.
Databricks Publish to Power BI: Your Measures Now Live in Unity Catalog, One Way Only by Ilya Rzhevskiy · 1 min read Databricks launches a one-way publishing feature syncing Unity Catalog metric views as DAX measures in Power BI semantic models.
v3.6.2 Release by Darren Gosbell · 2 min read The v3.6.2 update enhances Delta Analyzer's column mapping, improves Power BI Desktop discovery on Windows 11 with a thread-safe cache, and fixes editor and tracing issues.

Report Authoring and Interactivity

Power BI Gantt Chart: 7 Tips to Make It Useful by Valerie Junk · 1 min read Seven tips for making Power BI's native Gantt chart genuinely useful: group tasks, add milestones, show progress, use colour with purpose, shade weekends and use the zoom slider.
The Only SVG in Power BI Video You Need by Chandeep Chhabra Chandeep builds custom SVG tables and charts in Power BI for budget-versus-actuals and variance reporting, including range plots built with AI assistance, as a flexible alternative to native visuals.
What is Claude Data, and what does it mean for your semantic model? by Ruben Van de Voorde · 2 min read Claude Data adds certified queries, live sharing, version history and design-system support to Claude Dashboards, all running on your semantic model and its measures, so a better model leaves less room for invented KPIs. With no source control or pipeline deployment, the authors recommend promoting important dashboards to Power BI reports or Fabric Apps.
ChatGPT Codex + Power BI | Report Performance Check and Fix Slow Visuals by Pankaj Namekar Part 9 of the series runs Performance Analyzer on a slow report page (cards near 3,000 ms, YoY measures around 2,500 ms), digs into DAX query versus visual display time, then hands the findings to ChatGPT Codex to optimise slow measures and page design, re-testing after each change.
PBI is (not) Dead - Ep.579 - Power BI tips by Power BI Tips Mike and Tommy take on the perennial 'is Power BI dead?' question, debating whether AI and Fabric are replacing the tool or reshaping who uses it, and what that means for people who build BI for a living. News covers the retirement of R and Python visuals and Microsoft's sixth year as a Gartner data integration Leader.
Announcing the retirement of R and Python visuals in Power BI by Zoe Douglas · 2 min read Power BI's built-in R and Python visuals retire in May 2027: at that point they stop working in published reports (including older ones), R-powered custom visuals stop too, and new Desktop versions drop them. Python scripts in Power Query are unaffected. There's no in-place migration, so rebuild with built-in or custom visuals, or move the analysis to Fabric notebooks or Fabric Apps; PBIP users can search for pythonVisual and scriptVisual.

Deployment, Security and Operations

Power BI Pro vs F64 (1-0) - Why your small model can fail to refresh on Fabric by Dániel Gábor Patkós · 1 min read Small import models that refresh fine on Pro can hit 'not enough memory' on Fabric, even F64, because heavy Power Query work (late filtering, merges, sorts) drives 93–99% of peak refresh memory; one 75 MB model peaked at over 5 GB. File size doesn't predict refresh memory and a bigger SKU isn't the fix: refresh once on a large enough capacity, read the peak from workspace monitoring and compare it with the target SKU's per-model limit. Part 1 of a series.
Fabric Engines on 1B Rows: Pick by Query Shape by Sai Neelakantam · 1 min read Seven query shapes over about a billion synthetic Delta rows on a 64-CU Fabric trial, run cold and warm on Spark, the SQL analytics endpoint, a Direct Lake semantic model and Eventhouse. Direct Lake was fastest warm for aggregates and joins (about 0.6–0.8 s) but failed every cold run on four of seven shapes, hit the 10 GB per-query memory limit on a row-level top 100, used over ten times the SQL endpoint's estimated CUs and throttled the trial twice. Ends with a picker by query shape.
Publish with a pipeline: deploy it right (part 3) by Ruben Van de Voorde · 1 min read Part 3 of the series adds a Deploy stage after Validate that publishes the same commit to the workspace as a service principal, stamps the commit into deployed assets so you always know which version is live, and argues CI/CD is mostly about the discipline that only the pipeline deploys.
Org apps vs. Power BI apps: čím se liší a kdy je použít [Czech] by Jiří Neoral · 5 min read Compares org apps with workspace apps (formerly Power BI apps): org apps are Fabric items, allow several per workspace, propagate Read permissions to underlying models automatically (even across workspaces) and need no install, but have no versioning, so changes go live on Save. Workspace apps remain the choice for dashboards, scorecards, comments and mobile audience tabs, with a scenario table for deciding.
Set Up an Azure DevOps Repo: Power BI Project Files CI/CD by Ashish kumar Jha Video 3 of a Power BI with Git series connects an existing local Git repository holding a PBIP project to a new Azure DevOps repo and pushes it, laying the groundwork for a CI/CD pipeline (built in a later step).

General