FabCon and SQLCon 2026 brought nearly 5,000 data professionals to Barcelona, along with a long list of announcements. Arun Ulag’s summary on the Azure blog is a 15-minute read, and the product names come thick and fast. I’m building an AI-ready data lake on Microsoft Fabric, so I read it with one question in mind: what actually changes for the people building data platforms? Here’s my plain-language version.
The theme is simple: Copilot and agents are only as good as the business context they can reach, and Microsoft wants Fabric, now used by more than 40,000 customers, to be where that context lives.
The key takeaways
1. Fabric IQ is now part of Microsoft Copilot
Fabric IQ is Microsoft’s shared intelligence layer, combining OneLake data, trusted Power BI semantic models, and operational context from ontologies and Real-Time Intelligence. Fabric IQ in Copilot Chat and Cowork is now generally available, without additional AI token costs, with the new Code experience coming soon through the Frontier programme. People can ask Copilot business questions and get answers grounded in the semantic models their organisation already trusts.
2. Power BI moves from reports to apps
Power BI Desktop is getting agentic app creation. You start from a trusted semantic model, describe the application you need, then preview, edit and publish it to Fabric. These apps can accept inputs, write data back and support workflows, each backed by a SQL database in Fabric. Apps in Power BI will be in preview in the coming weeks, including for Pro and Premium Per User customers, who get Fabric Apps and Fabric Database capabilities, up to 1 GB per app, at no additional cost.
3. Fabric Apps grows up for production
Fabric Apps is the governed application layer for developers, who can use the open-source Rayfin SDK and CLI to build backends and deploy them to Fabric. Apps can now connect directly to warehouses, SQL databases, lakehouses and semantic models without duplicating data, while keeping existing permissions. TypeScript backend functions, a Secret Store, PostgreSQL support and private-by-default access fill in the production gaps.
4. OneLake becomes an open data network
Alongside mirroring and shortcuts, OneLake gains IQ sharing, in preview, for sharing governed data and business context with other teams, customers and partners. At launch it supports governed tables and files, Markdown agent instructions and RDF ontologies. There’s also a preview of bidirectional integration with Salesforce Data Cloud 360, general availability of Google BigQuery mirroring, and a growing ecosystem with Atlan, dbt and ClickHouse.
5. A data engineering agent for the long, hard jobs
The data engineering agent, in preview and built on technology from Microsoft’s acquisition of Osmos, targets long-running work such as migrations, lakehouse modernisation and complex ETL. Engineers define the outcome and guardrails; the agent plans, executes and validates within them. Tasks inherit workspace permissions, and the work can be steered from GitHub Copilot, Visual Studio Code, Codex or Claude Code.
6. Ontologies and agents get easier to build
A new agentic experience creates ontologies from semantic models, OneLake data and context documents. You can write rules in natural language and carry Power BI measures into the ontology, so established metrics are reused rather than redefined. Fabric data agents integrate more deeply with ontologies, and operations agents are adding pre-approved actions and root-cause analysis in preview.
7. Running the platform: observability, billing and databases
Observability in Fabric, in preview, brings workspace monitoring, the monitor hub and operations agents into one view. On-demand billing expands pay-for-what-you-use compute, and Fabric zero-provisioned removes the upfront capacity commitment. The Database Hub in Fabric, in preview, manages SQL Server, Azure SQL, PostgreSQL, Azure Cosmos DB and SQL database in Fabric from one place, and DiskANN vector indexes are now generally available in SQL database in Fabric.
What it means for data teams
- Semantic models are now AI infrastructure. Copilot answers, Power BI apps and ontologies all build on them, so names, descriptions and certified measures matter more than ever.
- Context is a deliverable. Ontologies, agent instructions and rules are things you design, govern and share, just like tables.
- Copy less, connect more. Mirroring, shortcuts and IQ sharing keep pushing towards one governed copy of the data.
- Engineers become supervisors of agents. Setting good outcomes and guardrails is turning into a core data engineering skill.
- Check the status labels. Much of this is in preview or “coming soon”, so plan pilots rather than production commitments for now.
My view
For the Fabric lake I’m building, this reinforces something I already believed: the hard part of AI-ready data isn’t the AI, it’s the shared meaning. If “active customer” or “net revenue” is defined three ways, no agent will fix that for you. I’m most interested in agentic ontology creation and the data engineering agent, and I’ll try both in a non-production workspace first. IQ sharing could be useful for sharing context with partner organisations, but I’d want to see the governance work in practice before relying on it.
If you only do one thing after FabCon, review your gold layer and semantic models as if an agent were going to read them tomorrow. Increasingly, one will.
Source
All facts in this post come from Arun Ulag’s article on the Microsoft Azure Blog, FabCon and SQLCon 2026 in Barcelona: Building the data foundation for Microsoft Copilot and agents (28 September 2026). Statuses change, so check Microsoft’s documentation before planning around them.
Leave a Reply