Eight of us went to Barcelona to separate what is ready for production from what is still roadmap. Here is the list, sorted into adopt now, watch, and leave alone for now.
Eight of us went to Barcelona to separate what is ready for production from what is still roadmap. Here is how we have sorted it for mid-market organisations on Power BI and Fabric.
Before FabCon we said we would sort what we heard into three piles: adopt now, watch, and leave alone for now. This is that list. It is written for organisations with a small data team, a Power BI estate they rely on, and no appetite for paying to be an early adopter.
The short version
- Adopt now: Fabric IQ in Copilot, after reviewing which semantic models Copilot can reach.
- Adopt now: a check of the capacity overage setting on every Fabric capacity you run.
- Watch: F0 and on-demand billing, Business Central mirroring, apps in Power BI, and the new route off Azure Data Factory.
- Leave alone for now: anything in early preview that would sit underneath production reporting.
Adopt now
Fabric IQ in Copilot
Fabric IQ is generally available. In Microsoft Copilot it chooses the relevant Power BI semantic model or report and answers from it. The Fabric IQ MCP server, also generally available, gives other agents the same governed access, from GitHub Copilot to agents you build yourself. Fabric data agents are generally available in Copilot Studio too, which puts them in Teams and Microsoft 365 Copilot.
This is the announcement we expected and the one that matters most. It confirms what we said before the event: AI agents are only as reliable as the business definitions beneath them. Your semantic model is now the thing your AI reads. If two models define revenue differently, Copilot will give two different answers with equal confidence.
Review which models Copilot can reach
Each semantic model has a setting that controls whether read-only users can reach it through Copilot, data agents and MCP servers. It is on by default. Before anyone switches Copilot on, list which models should be reachable and switch the setting off for the rest. Then check that the reachable models have clear names, descriptions and agreed definitions. It is an afternoon's work, and it is the cheapest AI governance you will ever do.
Check your capacity overage setting
Capacity overage is now generally available. Usage above your capacity is billed on a separate meter at three times the pay-as-you-go rate, and the threshold you set is a spending threshold, not a hard cap. Microsoft's documentation is not consistent on whether overage is on by default, so check every capacity you run. On-demand billing for Spark, the renamed Autoscale Billing for Spark, is also generally available, and worth switching on if notebooks are what spike your capacity.
Watch
F0 and on-demand billing
F0 is a zero-provisioned capacity: there is no compute to reserve before you start. On-demand billing extends to the other Fabric workloads, charged on actual use with a spend limit over a rolling 24 hours. For organisations that have held back from Fabric because of a monthly capacity commitment, this could be the change that matters most.
Both are arriving in preview, and Microsoft has not yet published F0 pricing, so we cannot yet say when F0 beats a small reserved capacity. We will update our Fabric Licensing Calculator when the prices are out.
Business Central mirroring
Business Central data can now mirror straight into OneLake. You choose companies and tables, including tables added by extensions, and they replicate as Delta tables within minutes, without touching Business Central's API limits. It needs Business Central 2026 release wave 2 and is in preview.
For our Business Central clients, this removes the extraction work that made Fabric a bigger step. It does not remove the modelling: Business Central tables still need shaping into something finance can trust, especially across several companies. We will run mirroring alongside existing pipelines before recommending anyone switch.
Apps in Power BI
Power BI can now build apps from a semantic model. You describe the app, preview it and publish it. Apps can take inputs and write data back, which covers the targets, forecasts and approvals that currently live in spreadsheets. It is in preview, and available to Power BI Pro and Premium Per User users, not only Fabric capacity customers. It is not yet suited to apps embedded for external customers. We are building with it now and will share what we find.
Moving off Azure Data Factory
Microsoft showed an upgrade path for Azure Data Factory: bring existing pipelines into a Fabric workspace, run an assessment that marks each one as ready, needs review or waiting on a planned capability, then upgrade them one at a time. A data engineering agent, in preview, is aimed at long-running migration and ETL work. Both make migration less of a cliff edge. Neither is a reason to move a stable estate before you have one.
Leave alone for now
Several things we saw are early previews: policies in Fabric, observability, GPU-accelerated warehouse queries and Database Hub. Some will matter. None should sit underneath the reports your leadership team relies on until they are generally available and have had a few months in real use.
Real-Time Intelligence remains the right tool for genuinely streaming operational data, and the wrong one for anything a daily refresh would serve.
What does not change
Every announcement points the same way. Copilot, agents, apps and mirroring all sit on top of the semantic model. The organisations that get value from this year's features will be the ones that already know what their numbers mean. If that is not you yet, the next step is not a new feature. It is agreeing your definitions.
If you want to know which of these apply to you, book a free audit.
Simon Devine
Managing Director
Part of the Hopton Analytics team, delivering governed analytics programmes for UK mid-market organisations.