May 12, 2026Microsoft Fabric, Podcast Episodes, Power BI

Episode 326 – Microsoft Fabric & Power BI April 2026 Feature Summary

Episode 326 is a double feature — and a first in a while. Recorded live and in person on May 7th, 2026 from Cologne, Germany at the European Collaboration Summit, John and Jason tackle both the Microsoft Fabric April 2026 Feature Summary and the Power BI April 2026 Feature Summary in a single session. The March Data Factory and Power BI content is on hold for a future episode — the hosts wanted to stay current as May approached — so this one picks up fresh with April and covers a lot of ground: DAX user-defined functions, Direct Lake getting calculated columns, true transactional DDL in Data Warehouse, VS Code deepening its grip on Fabric, and a raft of Real-Time Intelligence updates. April was lighter than recent months, which is how two feature summaries fit in one show.

A quick housekeeping note before diving in: the Fabric feature summary blog moved this month — it now lives on the community site at community.fabric.microsoft.com rather than blog.fabric.microsoft.com, which still redirects correctly. The new home has noticeably flatter formatting, especially in the table of contents, and the code samples embedded in the post need a careful look before copying. Worth knowing if you’re using the blog as a reference.

Fabric Platform: Tabs go GA, AI descriptions, and a Netezza farewell

The tabbed multitasking and Object Explorer experience is now generally available. Items open as tabs along the top of the Fabric experience — think Visual Studio-style tab management, not the Power BI experience (which doesn’t have this yet). Since preview, the update picked up right-click tab options, the ability to pop tabs out into a new browser window, a resizable Object Explorer pane, and support for monitoring jobs as tabs. It’s a meaningful quality-of-life shift for anyone managing multiple items at once. Two rough edges remain worth knowing about: clicking a workspace in the left rail currently opens a sub-workspace popup rather than navigating directly, and open tabs can shift far enough right to obscure admin icons like the tenant settings gear. Neither is a dealbreaker, but both are worth flagging before the feature surprises someone mid-demo.

AI Auto-Description for semantic models is now in preview. Point Copilot at a semantic model and it generates a plain-language summary based on the model’s metadata and structure — what’s in it, how it can be used. Model owners and contributors can generate, edit, and apply the description directly from the details page. It’s a Copilot-backed feature, which means it requires a Fabric capacity; it won’t show up in a trial or non-capacity environment. The Netezza built-in ODBC driver is also on its way out — the new driver went GA a few weeks back and deprecation of the old one has begun. No new connector install is needed; existing connectors work, but the ODBC driver itself requires an update.

Data Engineering: Notebook retry policy, VS Code deepens, Maven arrives

Notebook retry policy is one of the more practically useful additions this month. Fabric notebooks can now automatically restart after system errors — the canonical case being a Spark cluster getting recycled mid-run. The policy is configured via a %%configure block with a retriableOptions section, and a typical setup restarts up to three times with roughly 120 seconds between attempts. One important scope note: retry only applies to notebook jobs triggered via public API, Data Factory pipeline, or Eventstream — interactive runs in the UI are not covered. The blog post’s code sample had a formatting issue at publish time (the JSON structure wouldn’t work as written), though a Microsoft engineer corrected it in the comments shortly after.

VS Code integration with Fabric Data Engineering keeps expanding. The extension now lets you connect remotely to a Fabric workspace, browse workspaces and their Data Engineering items in the explorer view, edit notebook code locally with changes syncing back automatically, inspect and edit Environment items as YAML files, and set the default lakehouse for a notebook directly from VS Code. It’s not a code-first rethinking of the experience — it’s an alternate client that gives developers a familiar surface for things they’d otherwise have to do through the Fabric UI. For anyone doing the kind of source control and deployment pipeline work the hosts were presenting at ECS that same afternoon, the default lakehouse binding via VS Code is a direct workflow improvement. Maven support in Fabric Environments also landed in preview, letting Scala and Java developers manage library dependencies via pom.xml files rather than hunting down individual JARs to upload manually.

Data Science: ML gets OAP, cross-workspace logging, and SemPy 0.14.0

Machine Learning experiments and models can now be created and managed in workspaces with outbound access protection enabled — a gap that previously blocked data science teams in security-sensitive environments from using these capabilities at all. Cross-workspace MLflow logging extends that further: models and experiments can now be tracked across dev, test, and production workspaces using standard MLflow APIs, and assets from Azure Databricks, Azure Machine Learning, or any other MLflow-compatible platform can be imported directly into Fabric without rebuilding training pipelines. The synapseml-mlflow package is the on-ramp.

SemPy 0.14.0 is the headline for anyone doing programmatic Fabric administration from Python. The new sempy.fabric.admin module ships with 75 admin APIs covering workspaces, capacities, domains, tenant settings, reports, users, and more. The release also enables deploying semantic models across workspaces with automatic remapping of Direct Lake connections to new lakehouses or warehouses — a real CI/CD unlock for BI that the hosts have been watching for. Additional highlights include programmatic extraction and updating of Power BI report layouts, Delta OPTIMIZE on lakehouse tables from Python, and finer control over long-running operations via the new LroConfig. The name “semantic link” started as a gateway to semantic models; what it’s become is considerably broader.

Data Warehouse: True transactional DDL and JSONL ingestion

ALTER TABLE in explicit user transactions is now generally available — and it’s more significant than the name suggests. Previously, ALTER TABLE couldn’t participate in an explicit transaction even though other DDL and DML operations could. That meant schema migrations combining ALTER TABLE with other operations couldn’t be executed atomically, leaving the door open for partially applied schema changes on failure. With this GA, a BEGIN TRAN … COMMIT block can now wrap schema changes alongside data operations, making enterprise migrations safer and more predictable. COPY INTO for JSONL files also arrived this month — newline-delimited JSON is now a first-class FILE_TYPE option, which closes a gap for teams ingesting event streams, logs, and application exports that commonly arrive in JSONL format.

Real-Time Intelligence: CDC into Eventstreams, remote MCP, schema evolution, and more

The RTI section is busy this month. The new Mirrored Database Change Feed connector lets you stream row-level changes from mirrored databases directly into Fabric Eventstreams — inserts, updates, and deletes with full fidelity and source table structure preserved. It works with all mirrored database sources including Azure SQL, Cosmos DB, Oracle, PostgreSQL, Snowflake, and Open Mirroring partners, and it requires no custom Spark notebooks to poll for incremental updates. The change feed lands in an Eventstream where standard SQL operators, filtering, aggregation, and Activator can act on it. Two natural paths: route to Eventhouse for historical accumulation, or attach Activator and take action on the change itself without persisting anything.

Eventhouse remote MCP is now in preview — the hosted, cloud-based version of the MCP server that previously required a local install. Point GitHub Copilot, Copilot Studio, or Azure AI Foundry at the Eventhouse endpoint and AI agents can query real-time data, discover schemas, generate KQL, and sample data without an intermediary. A meaningful step for teams building real-time AI experiences. Eventhouse OneLake availability also picked up an important fix: you can now add or delete columns on a table with OneLake availability enabled without having to turn it off first. Previously that workflow risked data loss and required toggling the feature off and back on. It just works now.

Activator rule management moved into Eventstream itself — a “Set Alert” button in the ribbon lets you define conditions, configure actions, and create rules without leaving the Eventstream canvas. For teams that previously had to context-switch into the Activator UI to wire up alerts, this is a clean workflow improvement. Custom CA and mTLS support in Eventstream connectors also landed in preview, letting teams store their own certificates in Azure Key Vault for Kafka-based sources (Apache Kafka, AWS MSK, Confluent Cloud) and Confluent Schema Registry — solving the previous limitation where source systems with non-standard CA certificates simply wouldn’t connect. Eventstream observability via Workspace Monitoring also arrives in preview, emitting node status, throughput metrics, and error data into a monitoring Eventhouse as KQL tables. Worth noting the same architectural tradeoff that applies everywhere with workspace monitoring: the Event House standing up to monitor your Eventstreams is itself consuming capacity.

Power BI: Direct Lake gets calculated columns, DAX UDFs improve, and reporting gets polish

The biggest Power BI headline this month is Direct Lake calculated columns. Direct Lake on OneLake now supports calculated columns and calculated tables — in memory, so no storage materialization. The practical unblock here is significant: if you need a column in your model and don’t have access to upstream sources to add it there, you now have an option that didn’t exist before in Direct Lake. The user-context-aware capability enables translation scenarios — column headers that display in the user’s local language, for instance — which has been a persistent gap in Power BI’s multilingual story. Worth being precise about what this covers: it handles translation and locale-aware display, not currency conversion (which requires a calculation, not just user context metadata).

DAX user-defined functions picked up a small but useful enhancement: functions can now return specific elements of an object name — table name, model name, field name — individually, rather than returning the whole thing and parsing it out. DAX UDFs have been on a steady improvement trajectory and this closes one more friction point. On the reporting side, canvas settings now offer new aspect ratio options (the 16:9 fixed canvas isn’t the only choice anymore), visuals can be set to fixed size with scroll bars rather than dynamic resizing, and the card visual picked up category interactivity — click a card to cross-filter the page, similar to selecting a value in a slicer. The narrative visual UI changed so authors can switch between Copilot mode and the classic custom mode at any point during authoring rather than upfront. Preview visuals are now more obviously labeled in the visualization pane — they appear below the divider and carry a “preview” tooltip alongside the lightning bolt icon. The Azure Maps visual finally syncs the style you pick on the map canvas back to the properties panel, which has apparently not been the case until now.

Power BI Desktop is getting the updated file picker that’s been in the service for a while — the old one is being retired. Copilot in Power BI mobile picked up expanded capabilities for querying data on the go. Worth flagging: both the Fabric and Power BI summaries list the Netezza ODBC driver deprecation, which the hosts caught and called out as the same item appearing twice.

With the April summaries wrapped, the next stop is TechCon in Chicago in mid-June, followed by the European Power Platform Conference in Copenhagen. The March Data Factory content is still on the list — it’ll surface in a future episode when there’s space to give it a proper treatment.

Links

Microsoft Blog Posts

Previous Episodes

Events

Subscribe: SoundCloud | iTunes | Spotify | TuneIn | Amazon Music

Tags: Activator, ALTER TABLE, Azure Maps, calculated columns, Card Visual, Copilot, Data Engineering, Data Science, Data Warehouse, DAX, DAX user-defined functions, Direct Lake, Event Stream, Eventhouse, JSONL, Maven, MCP, Microsoft Fabric, Mirroring, ML experiments, narrative visual, Notebooks, Object Explorer, OneLake, Power BI, Real-Time Intelligence, retry policy, Semantic Link, SemPy, tabbed multitasking, VS Code