Power BI Weekly

Issue 378

9^th October 2026

No official announcements from the Power BI team this week (perhaps everyone is still recovering from Barcelona), so straight into the community content.

Last week I suggested keeping APPROXIMATEDISTINCTCOUNT away from your financials, and Chris Webb has put it through its paces on a 1.4 billion row Direct Lake model in The Problem With Approximate Distinct Counts In Power BI – And How To Solve It, suggesting a field parameter that lets users switch between the fast approximate count and the exact one (a sensible compromise, methinks).

Also following on from last week's feature summary, Sandeep Pawar has shown how to programmatically control Copilot access for semantic models from a Fabric notebook, which is handy given the setting is on by default (though note it relies on an internal API that could change). Sandeep has also written about Fabric-RLM: Working With Semantic Models. Sandeep created Fabric-RLM as a Python package for Fabric notebooks where you describe a task, its inputs, the output you expect and the LLM to use, and it plans, runs code and fixes its own errors in a single loop. In this post, he uses a Power BI semantic model as the input to build HTML reports that check their own numbers before handing them over.

Finally, John Kerski has introduced code coverage for semantic models in pql-test, showing which parts of your model your DAX tests actually touch (while noting that "coverage" isn't the same as "correctness").

Data Prep

Power Query Has Changed Forever: the Good and the BAD by Chandeep Chhabra Chandeep Chhabra walks through the redesigned Power Query editor in the September 2026 Power BI Desktop update. He covers collapsible panels, better Applied Steps and query folding indicators, the new Data, Schema, Diagram and Script views, and Rank Column and Table.ClearDown. He also flags regressions for M authors: no leading equals sign in the formula bar, changed zoom, and much more limited IntelliSense.
Fabric Dataflow Gen 2 - Upgrade Accelerator by Pat Mahoney Pat Mahoney walks through the new UI for upgrading Power BI (Gen1) dataflows to Fabric Dataflow Gen2 (CI/CD). He also shares a notebook that uses two new REST APIs to check upgrade readiness and upgrade dataflows across multiple workspaces.
PowerQuery UI Changes - Sep 2026 Power BI Update by Vijay Perepa The Sep 2026 Power BI update introduces a redesigned Power Query Editor UI, which this tutorial explains to help users quickly adapt despite the new layout while preserving core functionality.

Data Modeling

Understanding DAX contracts when using AI by Alberto Ferrari · 1 min read A contract is a short business-language document defining how a semantic model computes its values, and SQLBI argues it should come before any measure when writing DAX with AI. Using a customer-ranking example, the article (with companion video) shows the agent asking clarifying questions, the contract being refined over a few prompts, and only then the measure generated from it. The finished contract is far easier to read than the DAX it produces.
The Problem With Approximate Distinct Counts In Power BI – And How To Solve It by Chris Webb · 2 min read Testing ApproximateDistinctCount() on a 1.4 billion row Direct Lake model, Chris Webb saw a query drop from about 6 seconds and 82 seconds of CPU to about 4 seconds and 48 seconds, at the cost of more memory and results roughly 1.6% off. Since users may reject approximate numbers, he suggests a field parameter that defaults to the fast approximate count but lets them switch to the exact one.
Fabric-RLM : Working With Semantic Models by Sandeep Pawar · 3 min read Sandeep Pawar binds a Power BI semantic model into Fabric-RLM from a Fabric notebook. He uses it first to answer a question via schema inspection and self-corrected DAX, then to generate a full HTML report with charts. Validation functions independently recompute the headline figures and send the report back for repair if they don't match.
Who Is in the Loop? Semantic Models, Meaning and People by Juliana Smith · 7 min read Reflecting on FabCon sessions about semantic models and Fabric data agents, Juliana Smith argues that most wrong AI answers are disambiguation failures, so models must make business meaning explicit through names, descriptions, synonyms and definitions. She also covers testing agents against a ground-truth question set and inspecting the DAX they generate, keeping people at the centre with AI "in the loop".
ChatGPT Astra + Power BI | Data Modeling and Documentation | Part 3 by Pankaj Namekar Part 3 of Pankaj Namekar's ChatGPT + Power BI series uses ChatGPT Codex to design a star schema, pick relationship cardinality and direction, validate the model before any DAX is written, and generate documentation for the finished model.
ChatGPT Astra + Power BI | DAX Deployment Done Right | Part 4 by Pankaj Namekar Part 4 of Pankaj Namekar's ChatGPT + Power BI series uses ChatGPT to write, debug, comment and organise DAX measures (CALCULATE, FILTER, ALL, RELATED and time intelligence), then walks through deploying them into a Power BI model with a consistent naming convention.

Report Authoring and Interactivity

Copy Selected Cells from Power BI to Excel (Custom Visual) by Maciej Krakowski · 1 min read Maciej Krakowski demonstrates Advanced Grid Table, a free community custom visual he developed. It lets report users select a range of cells and copy them straight into Excel, which helps when you need a few values for further analysis or a quick check during a meeting. The visual and a sample report are available on GitHub.
Power BI Data in Excel: Export to Excel vs Analyze in Excel (Why One Wins) by Reza Rad Analyzing in Excel offers live data updates, respects row-level security, and bypasses export limitations, making it the superior choice over Export to Excel for working with Power BI data.
How to Create a Variance Line Chart in Power BI Using Native Features by Kate Dashboards This Power BI tutorial shows how to create a dynamic Variance Line Chart with colored gap lines, auto-updating labels, and a Measure Selector for switching financial metrics without custom visuals.

Deployment, Security and Operations

#7 Planning Your Fabric Rollout: Pilot vs. Enterprise-Wide Adoption Strategies by Paul Turley · 1 min read Part 7 of Paul Turley's Fabric getting-started series argues that pilot versus enterprise-wide is the wrong debate: what matters is which one your organisation is ready for. He recommends starting with an isolated proof of concept, then a pilot covering roughly 10–15% of eventual users to surface governance and skill gaps. An existing mature Power BI footprint can justify going wide sooner. Either way, define success and exit criteria up front, name stakeholders early, and add workloads only as governance maturity catches up.
Code Coverage for Semantic Models: Introducing pql-test code-coverage by John Kerski · 4 min read pql-test 0.1.18 adds a code-coverage command for semantic models. It counts a table, column, measure, relationship or role as covered only when a DAX test UDF references it directly (first-hop edges from INFO.CALCDEPENDENCY), and it can fail a run below a minimum threshold. John Kerski is explicit that coverage shows what is referenced by a test, not that the assertions are meaningful.
Who’s Viewing Your Power BI Reports? Power BI Usage Metrics by Ashish kumar Jha A walkthrough of Usage Metrics in the Power BI Service: creating the report, then reading views, unique viewers, page usage, trends, distribution methods, platforms and performance to see which reports are actually being used.
Programmatically Control Copilot access for semantic models by Sandeep Pawar · 1 min read A new semantic model setting, on by default, decides whether read-only users can reach the model through Copilot, report Copilot or data agents. Anyone with write permission can switch it off in settings. Sandeep Pawar shows how to read or toggle it from a Fabric notebook with SemPy, noting that this uses an internal API that may change.
The Hidden Security Risks of Role Combination in SSAS Tabular by Pablo Echeverria · 5 min read When a user belongs to several SSAS Tabular roles, the engine evaluates each role separately and unions the visible rows, so one unrestricted role silently overrides every RLS filter. Pablo Echeverria provides a PowerShell script to audit roles, members and filters. He shows how to test with Roles and EffectiveUserName in the connection string and recommends putting a user's intended restrictions into a single role.

General

Arun Ulag on the Future of Fabric, Power BI and AI | Fabric Insider Ep. 23 by Reza Rad At FabCon Europe, Reza Rad puts the community's questions to Microsoft EVP Arun Ulag. Topics include why expand from Power BI into Fabric, how Fabric compares with Databricks and Snowflake, whether Fabric Apps replace Power BI, Power BI's identity now that AI can generate reports, where Fabric IQ is heading, and career advice for the AI era.
Where Writeback belongs - Ep.569 - Power BI tips by Power BI Tips Mike Carlo and Tommy Puglia debate whether Power BI should stay an analytics layer or become an operational one. They cover when write-back belongs in a report, who owns data once users can edit it, and what governance needs to come first. As AI surfaces signals faster, they argue, the bottleneck shifts to the disconnected workflow after the dashboard.
FabCon Barcelona 2026 Recap by Eugene Meidinger · 1 min read Eugene Meidinger's FabCon Barcelona takeaways: Fabric Apps are coming in preview to Pro and PPU licences, and Microsoft has joined Apache Ossie for semantic model portability. Ontologies can now be built from multiple semantic models with DAX measures as metrics. PBIP and the Power BI agentic experiences (authoring skills, authoring MCP server and Desktop Bridge) are now GA.
Mourning the (alleged) death of the Power BI Developer and thriving in uncertain times by Eugene Meidinger · 9 min read Eugene Meidinger makes room to grieve amid "Power BI is dead" talk. He argues that Power BI reporting may drift the way SSRS did, while semantic modeling and DAX have a healthy future even as AI writes more of the code. His advice is to diversify your skills and invest in accountability-heavy work, your own judgement, regular practice with frontier AI, Vega-Lite, data modeling and requirements gathering.