The first day of FabCon is always a bit of an information overload.
This year in Barcelona wasn't any different. The keynote alone jumped from Fabric IQ integration with Microsoft 365 Copilot to on-demand Fabric compute, IQ sharing and new Power BI experiences and what felt like another announcement every few minutes. At some point, taking notes starts to feel less like documentation and more like competitive typing.
Luckily, the corenotes and breakout sessions gave us a chance to digest individual areas in more detail. Microsoft also published a hefty amount of material during the conference, so there was no shortage of reading for the evenings.
I have developed another habit during FabCon as well. We keep an internal live feed at Zure where I try to share the most interesting announcements as they happen, together with a quick interpretation of what they might actually mean for our day-to-day work.
If someone at Zure isn't sitting in the same room, the goal is that they can still follow what is happening, challenge the announcements and start thinking about where they might matter in our customer work.
So after several days of messages, screenshots and “okay, this one is actually interesting”, here are a few things that caught my attention.
Fabric wants to understand your business
Fabric IQ was difficult to miss.
Microsoft announced its availability through Microsoft 365 Copilot Chat and Cowork, and it appeared repeatedly in keynotes and sessions. Fabric IQ is one of the four parts of Microsoft IQ together with Work IQ, Foundry IQ and Web IQ. Very roughly: Work IQ understands how people work, Foundry IQ brings organizational knowledge, Web IQ brings outside-world context, and Fabric IQ provides the business context around your data.
That last part is important.
For years, organizations have invested in semantic models containing measures, relationships and business definitions. Fabric IQ makes that foundation available beyond Power BI reports and gives people and agents access to the same context.
The demo made this look wonderfully simple. Ask something in business language, let Fabric IQ understand what a customer, margin or product actually means, and use that context in Copilot. More importantly, the answer doesn't appear from a black box: the user can see which report, semantic model or KPI the answer is grounded in.
There is, naturally, one tiny prerequisite: the context needs to be good.
If your semantic models contain mysterious measures called Measure 7, conflicting definitions of revenue and descriptions last updated during COVID lockdowns, an AI agent isn't going to magically turn that into a pristine business language layer.
No biggie.
Jokes aside, I think this changes the importance of semantic modelling quite a bit. Good measures, relationships, descriptions and ownership used to be mainly about keeping analytics understandable and consistent for humans. Now that same business context increasingly needs to be understandable by machines as well.
If Fabric IQ is going to use existing semantic models to ground Copilot and agents, semantic-model hygiene is no longer just BI housekeeping. It starts becoming part of your AI architecture.
Microsoft also gave an early look at semantic views. The idea is to bring governed business metrics and semantics closer to where the data lives in OneLake and reuse that context across analytics, applications and AI. Microsoft's original pitch was to bring all your data into OneLake. Now the next step seems to be bringing the meaning of that data there as well.
That is particularly interesting to me. Today, a lot of our carefully defined business meaning lives inside semantic models. Semantic views suggest that some of that context could also be defined closer to the underlying data itself, making it reusable by more than just Power BI.
There are still plenty of open questions around how semantic views and existing Power BI semantic models will eventually work together, so this one deserves its own deeper look later. But one thing seems fairly clear: all the data-platform work you've already done isn't suddenly becoming obsolete because AI showed up. Giving the data proper meaning might actually become more important than ever.
Fabric without remembering to press Pause
Another announcement that immediately caught my attention was Fabric's zero-provisioned F0 capacity, together with broader on-demand billing.
Anyone who has used a small Fabric capacity for development probably knows the routine. Start the F2 in the morning. Do some work. Remember to pause it in the evening.
At least that was the plan.
F0 changes this model. Instead of provisioning compute upfront, supported workloads can consume compute on demand. That makes a lot of sense for experimentation, development and workloads whose consumption is difficult to predict.
Of course, “pay for whatever you consume” raises the obvious follow-up question: can I now accidentally consume a lot?
Potentially, yes. Fortunately, Microsoft is introducing guardrails for on-demand consumption as well, so administrators can put spending limits around different types of usage before new operations are paused.
So we get elasticity without necessarily giving every experimental notebook an unlimited corporate credit card.
The same lowering of entry barriers can be seen on the Power BI side. The upcoming app-building experience in Power BI will also be available for Power BI Pro and Premium Per User customers, with Fabric Apps and Fabric Database capabilities included within defined limits.
That is an interesting shift. Power BI development has traditionally ended with reports and dashboards. Fabric Apps pushes that boundary toward interactive applications that can accept input, persist state and participate in actual business processes.
Fabric Apps are built on Rayfin underneath, so I suspect we'll see quite a few new application ideas appearing in Fabric workspaces soon.
The boring things are getting better too
Not everything in Barcelona was about agents.
Platform observability, monitoring and governance received a significant amount of attention, and I was happy to see it. These aren't necessarily the features that fill a keynote screenshot, but they become increasingly important once Fabric grows beyond a handful of workspaces.
The refreshed Monitor Hub brings capacity information and operational actions closer together. Microsoft is also expanding operations agents that can investigate failures, gather supporting evidence and suggest next actions. The current documented scope includes investigating pipeline failures, with more scenarios planned.
Governance is becoming more granular as well. New Fabric Policies introduce centralized, attribute-based controls for scenarios such as item creation and workspace security. That's a welcome direction compared with relying only on broad tenant-level switches for an increasingly large Fabric estate.
For me, these announcements are a sign of Fabric growing up. Creating things has never been the hardest part of a data platform. Running hundreds of them consistently, understanding what broke and controlling who is allowed to do what is where things become interesting.
And then there are the small things
These are only a few announcements. GPU-accelerated Fabric Data Warehouse probably deserves an article of its own. Project Osmos raises interesting questions about how much of data engineering can eventually be delegated to agents. And the continued evolution of semantic models deserves much more space than I have here.
It is fascinating to compare the platform today with what we started using during the 2023 public preview.
Still, among all the Copilots, GPUs, agents and IQs, some of the most satisfying announcements are often much smaller.
Pipeline schedules, for example, can now pass parameter values from Fabric Variable Libraries. That is hardly going to make the keynote highlight reel, but our team spotted it immediately because it solved a very real problem for us — followed by a collective sigh of relief.
And sometimes that is exactly the feature you wanted.
What a wonderful world.
