 |
| FabCon Strolling DJ, in case you weren't sure whether All Star still had legs |
I attended FabCon 2026 in Atlanta two weeks ago. What follows are my scattered notes and reactions, organized alphabetically by topic — not because I love the alphabet but because it enables you to quickly scan for what's relevant to you.
Before the topics, a note on how I choose sessions at conferences like this, because it significantly affects what you get out of them: I generally skip sponsor presenters (who are presenting in order to sell you something) and customers (whose stories tend to be too thin on details to be useful). I like MVPs best, followed by Microsoft employees. MVPs can be more direct and brutally honest. There are exceptions, but that's my default approach based on my role and why I'm there — looking for things I can use to help my clients.
AI and Copilot
- Use case example: Use Copilot in DAX Query View to create dynamic SVGs in lieu of custom visuals (here's an example)
- Use case example: custom visuals and theme modification in VS Code with GitHub Copilot
- Operations Agent in public preview: similar to creating shortcuts in Claude with simple markdown, but continuously monitoring your data (some documentation)
- Microsoft says they're working on more agent-friendly report development, not just semantic models, which already have a lot of options (TMDL, Tabular Editor, Power BI Modeling MCP...)
DirectLake
- Composite models are supported in DirectLake, so not everything must be in the Lakehouse, per se
- Even in DirectLake models, you can change storage modes for individual tables to import if you like, which improves open time and enables you to transform data. The presenters suggested perhaps doing this for dimension tables, similar to using mixed storage for dimensions in DQ+import composite models.
- Composite models in DirectLake use "strong relationships" — unlike typical composite models when relating tables in different storage modes generate less efficient queries
- Calculated columns are coming for DirectLake. Beyond that, Microsoft is introducing expression context where a calculated column may recalculate at times other than when the semantic model refreshes, like when each user logs in.
- V-order optimization has some overhead, and it doesn't ALWAYS make sense to enable it. V-order benefit for a table = reads / writes. DAX queries are reads, so apply V-Order on the gold layer.
- Partition fact tables by calendar key in the lakehouse — creates a lot of subfolders but significantly shrink any query that scans that column
- Column cardinality (number of distinct values) has biggest impact on cache warming time
- Don't overengineer models that don't need it!
Enterprise Planning
- Built-in Enterprise Planning platform as a Fabric workload, included with the SKU, covering a lot of common writeback scenarios. I wouldn't be surprised if they locked this behind the F64 or higher SKU, else you could get an enterprise planning tool for a very small cost per month with an F2, even if you don't use Fabric for anything else.
Fabric costs
- There are new Workspace controls that enable you to put a ceiling on how much of your capacity a Workspace can use.
- Microsoft also solved a problem with people having mission critical Workspaces throttled to the point of being unusable by finding a way to let you pay them more. Ooh!
Mirroring
- Mirroring of Excel and SharePoint lists to Fabric is coming, and it means one fewer thing to schedule and administer when it comes to getting business user-managed data to the Lakehouse.
OneLake security
- Not as intimidating as it sounds — similar in many respects to Power BI roles and security, but with additional options for column-level, table-level, and folder/schema scope.
- Groups still need to be added to the workspace for read access to the lakehouse — security only restricts data; it still must be granted.
- Workspace security supersedes OneLake security if you grant a higher role than Viewer, similar to Power BI
- Hardcoded filters only, not dynamic — and no joins in the role definition. That's a big limitation if this is the only way you can apply RLS efficiently to DirectLake models.
- You can't apply row or column level security to shortcutted data — only where it resides
- No data masking yet
Paginated reports
- Amazingly bad product! No wonder most people avoid these. Somebody from Microsoft should be forced to watch this session, as the presenter worked around idiosyncrasies with paginated report development.
Power BI UI and visuals
- Custom totals for table/matrix (ex., sum of rows, count of rows). I poked around with these a bit and will write about them more soon.
- Revamped date picker
- Gantt as a native visual (based on the current Microsoft-build custom visual)
- One of my favorite things was an improved workflow for updating theme files. I find updating theme files to be a lengthy trial and error process, searching for properties, and endless iterations, but they showed changing the settings in an object and then pushing that object's settings to make them default in the current theme.
- There are some new kind of theme switcher, new default canvas sizes, and modern visual defaults. I'm lumping these together because I didn't quite understand what they were or all the implications based on the demos. For example, what happens with existing custom themes when you change these settings? What happens when you change the canvas size to the x and y coordinates of existing objects? Haven't played around with them yet.
TMDL
- TMDL view in the service coming soon (mentioned in multiple sessions — sounds like a few weeks out)
Whoopie cushions
- If you have young children and are going on a work trip, I highly recommend ordering whoopie cushions and timing them to arrive while you are gone.