Fabric Apps

Tommy Puglia, Mike Carlo, and Kurt Buhler spend this episode on what is left of the Power BI identity once Microsoft Fabric and agents are part of the job. The skill that still decides the work is judgment: listening to the business, and staying critical of what an agent hands back.

Power Query is moving a set of cloud connectors from ODBC to ADBC, and legacy Power BI Q&A now stays through February 2027. Kurt Buhler joins Tommy Puglia and Mike Carlo to sort out what data literacy and AI literacy require once Microsoft Fabric and agents make it easy to build faster than a team can review.

Fabric Apps can own every pixel of a data experience, and Kurt Buhler is not ready to hand them the executive dashboard. Tommy Puglia and Mike Carlo spend the hour on where a Rayfin app fits beside Power BI, and on the design system, the token bill, and the research bench a team needs before it scales.

Tommy Puglia and Mike Carlo spend this episode on how a Microsoft Fabric project moves to a team that has to run it. Their model is I build, we build, then you build, and it holds when leadership, trust, and the team's own context stay in place after the builder steps back.

Tommy Puglia and Mike Carlo read Rituparna Das's Nightingale guide to using AI in data visualization without an enterprise license, and they ask what those three workflows become when the output is a Power BI report. They also take apart a Fabric warehouse medallion best-practices post and a private-link path for copying data to Snowflake.

Foundry can now find Fabric data agents in the OneLake catalog and attach more than one of them to a single agent. Tommy Puglia and Mike Carlo walk through what that changes, from the 15,000-character instruction limit to a CU bill that still deserves a hard look.

On Explicit Measures, Tommy and Mike explain liquid clustering for Delta Lake tables in Microsoft Fabric and why Fabric Runtime 2.0 changes the cost of keeping that layout healthy. They walk Miles Cole's incremental optimize example and Mike's rules for when a Lakehouse Spark job is the right transform.

Tommy Puglia and Mike Carlo spend this episode on Microsoft's new guidance for choosing a medallion pattern in the Fabric Data Warehouse. Bronze, silver, and gold still describe the work; the open question is whether a warehouse belongs in every layer.

Eva Sibinga's Nightingale essay argues that large language models hallucinate by design and are sold to look helpful before they are right. Tommy Puglia and Mike Carlo take that diagnosis into Power BI and Microsoft Fabric: a harness of MCP servers, skills, and validated context, plus a prototype refined in stages, is how you keep the visual honest.

Anonymous data access lets a Rayfin app take a public write — a contact form, a game score, an outage report — and store it in a Fabric SQL database without an Entra account. Mike ranks that capability third: the case for Fabric apps is still that the data stays in Fabric, and that an agent can build and fix the app.

Mike has been building apps on Fabric as a backend since January, and he's coining it FAB. The case isn't that you already own Fabric — it's that the workspace identity, git integration, and SQL database remove the wiring you'd otherwise do by hand in Azure.

Mike is deleting paid subscriptions and rebuilding them as Fabric apps; Tommy wants to know who maintains all this and whether enterprise security teams will ever allow it. Plus the licensing discovery of the episode — agents talking to semantic models over XMLA with no Power BI Pro license required.