Last week I was in Barcelona for FabCon and SQLCon. Over 130 sessions made for a long announcement list, so I won’t repeat all of it! This is what stood out to me as a data engineer in the Managed Data & Insights team, looking after Fabric solutions day to day.
Observability
Monitoring and observability have traditionally not been a strong point in Fabric, but the announcements at FabCon were a step in the right direction. Unified workspace monitoring lets you send telemetry from multiple workspaces into one shared Eventhouse. Alongside it comes a refreshed Monitor Hub, improved pipeline monitoring, Activator-powered alerts for job successes and failures, and even an operations agent that investigates pipeline failures. I’m most keen to try the new Monitor Hub, as well as assessing the usefulness of the operations agent.
Smart retry
Pipeline activities now support retry back-off. Delays grow between attempts, within minimum and maximum bounds, rather than staying fixed. This will help absorb transient issues, and the “failed overnight, succeeded on re-run” tickets we see more regularly than any of us would like! Whilst this may seem like a small setting, I can see how this could have a direct impact on my day-to-day.
Runtime 2.0
Runtime 2.0 is now generally available, bringing Spark 4.1, Delta Lake 4.2, Python 3.13 and Java 21, plus a stronger Native Execution Engine. Microsoft quotes up to six times faster than open-source Spark on TPC-DS (which is definitely a best-case scenario!). Significantly, Delta Lake 4.2 brings incremental liquid clustering, meaning there is no longer the requirement for whole table rewrites when optimising. We’ve already started having conversations with clients around the best approach for upgrading from 1.3.
Project Osmos
Perhaps my favourite session was “From Prompt to Finished Project: Introducing Project Osmos – Data engineering agent in Microsoft Fabric”, on the new data engineering agent built on the Osmos acquisition. You set the outcome and the guardrails, and it plans, executes and validates the work on Fabric Spark. Tasks inherit workspace permissions, so colleagues can follow along, and you can drive it from GitHub Copilot, VS Code or Claude Code. The demo shown during the session was genuinely impressive, with the agent working at a whole project level rather than being task-based. The differentiation from when you would use this over standard Copilot with Fabric skills was clear – Osmos would spend a significantly longer time working (the demos were sped up over 4+ hours), but the quality is incomparable. The ability to customise with your own skills to instil you or your company’s style guide and best practices was also mentioned.
Zero provisioned F0 SKU
F0 is a newly announced zero-provisioned entry point with on-demand billing on by default, arriving in the coming weeks. Pricing for on-demand and F0 hasn’t been published, so I wouldn’t change any reservations yet. However, there are already some very clear use cases which we’ve all probably experienced in the past. Firstly, the age-old question: what size capacity do I need? Estimates of this are always just that – estimates. Now we have the option of starting on an F0 SKU initially to get a baseline, before deciding on the correct size and purchasing a reservation. Secondly, the noisy neighbour problem. We’ve all (unfortunately) experienced a long-running notebook throttling your capacity due to an unexpectedly large data volume. Now we have another potential solution in our arsenal – putting the notebooks in a separate workspace on an F0 capacity to isolate them and only pay for what we use.
Fabric Apps
Fabric Apps may have launched in June, but this FabCon demonstrated how powerful a tool they can be. Apps can now connect directly to warehouses, SQL databases, lakehouses and Power BI semantic models without copying data, with existing permissions carried through. New additions include TypeScript backend functions, a secret store, PostgreSQL support, and built-in usage and performance metrics. Power BI Desktop is also getting agentic app creation, where each app gets its own SQL database in Fabric. Excitingly, it’s also coming to Pro and Premium Per User customers as well as Fabric capacity customers. Given my role in the Managed Data & Insights team, the key question I took away is who supports an app once the person who built it has moved on. The usage metrics and private-by-default access are a good start, and the fact that apps inherit existing permissions makes it more important that workspace and lakehouse security is already in good shape.
The full list of announcements can be found here:
Fabric September 2026 Feature Summary | Microsoft Fabric Community
To summarise my thoughts, Barcelona felt less about big new ideas and more about making Fabric easier to operate, which for many of us is a win.
I’d highly recommend attending FabCon in the future, with the upcoming events being FabCon US in Atlanta (March 2027) and FabCon Europe in Amsterdam (October 2027).