FabCon Europe 2026: Fabric IQ in Microsoft 365 Copilot, ontologies and control over agents

A year ago, agent announcements were about tools: what an agent could call. This year they are about context: what an agent knows about the business before it calls anything. That is the right shift. An agent with a hundred tools and no idea what "active customer" means will use all hundred to hand you the wrong number.

Every major data platform now ships its own answer. Microsoft has Fabric IQ and its ontology, Databricks has Genie Ontology and metric views, Snowflake has semantic views, Google has Knowledge Catalog and Looker, and Oracle annotates the database itself.

FabCon and SQLCon Europe ran in Barcelona from September 28 to October 1, 2026, and Microsoft announced more in this area than at any event before it. Below is all of it, grouped by theme, with the release stage Microsoft gave each item, and then where everyone else stands.

Microsoft IQ

Microsoft now groups its context services under one name, Microsoft IQ, with four parts:

Web IQ, context from the public web.
Work IQ, context from Microsoft 365: mail, meetings, documents and people.
Foundry IQ, enterprise knowledge retrieval for agents built in Microsoft Foundry.
Fabric IQ, business entities, relationships, rules and live data from Microsoft Fabric.

Yitzhak Kesselman, the corporate vice president who leads Fabric's real-time and IQ work, set the theme in his posts that week. In his view, the challenge "isn't deploying more agents." It is giving agents a shared understanding of the organization, and he argued that the advantage will come from the quality of the foundation under the AI rather than the AI itself. In his recap he singled out one moment: demonstrating the Ontology Agent on the keynote stage, with business entities, relationships, rules and actions forming a model that people and agents both reason over.

Customers on stage made the same case. Sonepar described using Microsoft IQ to cut a sales notification step from about 45 minutes to under a minute, the speakers said. SPIE showed a digital twin of its buildings on Fabric IQ, used to find faults earlier, and said it had used Fabric IQ as an early-access partner before general availability. Microsoft said Fabric now has more than 40,000 customers and named Levi Strauss, H&M and Thomson Reuters among them.

Fabric IQ reaches Copilot

Fabric IQ in Microsoft 365 Copilot Chat and Cowork, generally available and on by default, with no added AI token cost. Copilot answers now draw on governed Power BI reports and semantic models in Fabric.
Answers from data agents and ontologies in Copilot, in preview, so Copilot can reach past reports and semantic models.
IQ sharing, announced for preview. It shares governed data together with its business context across teams, customers and partners. At launch it carries tables, files, Markdown instructions and RDF ontologies; native Fabric ontologies and semantic models are planned.
Fabric IQ in the new Code experience, coming soon through Microsoft's Frontier program.
OneLake catalog in Foundry, in preview, so agents built in Foundry can find Fabric data.

What's new in the ontology

The ontology item itself is still in preview, as Microsoft Learn states. Microsoft's posts gave no release stage for most of the features below.

Ontology Agent, in preview. It drafts an ontology from semantic models, OneLake data and documents, working through a chat in plain language, and helps govern it as it changes.
Semantic models as a starting point. The agent can generate an ontology from a Power BI semantic model and carry its measures in as metrics. During the preview the agent does not bind entities to the semantic model itself. A person can still bind an entity to a semantic model by hand, with Read and Build permission on it.
Bindings to mirrored databases and SQL databases, without copying the data. A table mirrored into OneLake from another platform can back an ontology entity.
Modeling features: namespaces, inheritance, keyless bindings and joinless relationships.
Lifecycle: named versions without Git, CI/CD support, management APIs and an Ontology Management SDK.
RDF and OWL import and export, so an ontology can move in and out of Fabric in a standard format.
Ontology Rules, business rules written in natural language.
Optional graph materialization, so the graph is built only when a workload needs it.
Ontology tools over MCP, so agents outside Fabric can query the ontology.
Security carried through. OneLake security (row, object and column level), outbound access protection and customer-managed keys apply to the ontology.
The ontology as context elsewhere, in preview: data agents can ground their queries in it, and Real-Time Dashboards can use it as a source.
Planned: integrations with Foundry IQ, Copilot Studio and the operations agent, and Fabric IQ for Data Warehouse.

New agents

Data agents in Copilot Studio, generally available (announced in early September and restated in Barcelona).
Data engineering agent, in preview. Built on the Osmos acquisition, it plans and runs migrations, ETL and performance work within guardrails engineers set. Tasks inherit workspace permissions, and engineers can start and steer them from GitHub Copilot, VS Code, Codex or Claude Code.
Operations agent updates, in preview: pre-approved actions, root cause analysis through a new Investigate experience, and a performance view.
Database agents, covered under SQL below.
Eventhouse onboarding agent, which sets up real-time analytics from a description of the data.
Agentic Copilot in notebooks and an AI mode for building agents that draws on semantic models, documents, glossaries and data dictionaries.
Fabric IQ MCP server, generally available. It covers Power BI reports and semantic models and enforces row and object level security. Ontologies and data agents are next.
Fabric Core MCP server, generally available, giving agents authenticated and audited access to workspaces, items, permissions, search and capacity information.
Data agent updates: charts in answers and long-running MCP tasks, both generally available, and service principal sign-in, in preview, so custom apps and Foundry agents can call a data agent with an application identity.
More MCP servers and skills: an Eventhouse MCP server, Data Factory MCP tools and skills, and an open repository of Fabric skills for coding agents on GitHub.

Power BI and Fabric Apps

This was the change with the widest reach, because of who gets it.

Apps in Power BI, preview in the coming weeks. In Power BI Desktop you start from a semantic model, describe the app you want, and publish it to Fabric. The apps accept input, write data back and keep state, and each one gets a SQL database in Fabric.
Included with Power BI Pro and Premium Per User, at no added cost, along with Fabric capacity customers. Pro and PPU users get Fabric Apps and Fabric database capabilities up to 1 GB per app. This puts application building in front of the people who build Power BI reports today.
Fabric Apps, in preview since June. Developers and coding agents define an app's backend in code with Rayfin, an open-source TypeScript SDK and CLI: data models, APIs, identity, access policies and business logic. The latest additions, all in preview: TypeScript backend functions, a secret store, PostgreSQL support, private-by-default access, built-in usage metrics and anonymous access to selected data for public-facing apps.

OneLake and mirroring

Salesforce Data 360, bidirectional integration in preview, so Salesforce data joins OneLake without copying and can serve as business context for data agents.
Google BigQuery mirroring, generally available, and deeper integration with Dynamics 365 Business Central.
Extended mirroring, generally available, with change feeds and mirroring of source views. Mirrored change feeds can also stream into Eventstreams, in preview.
Snowflake security roles in mirroring, in preview. Snowflake role definitions come into Fabric with the mirrored data.
SharePoint. SharePoint List mirroring and SharePoint folders in Copy job are generally available, and a site picker for SharePoint sources and Lakehouse shortcuts is in preview.
Delegated OneLake shortcuts, in preview. A shortcut can carry an organizational account, service principal or workspace identity, including across Fabric tenants.
Open access. The OneLake table read API (Delta and Iceberg), the catalog search API and the Fabric Core MCP server give other engines and agents a governed way in. Iceberg catalogs from AWS Glue, Dremio and Google are listed as mirroring sources in preview.
IQ sharing partners. EY, S&P Global, TomTom, Financial Fabric and xMentium are building integrations with IQ sharing, and TomTom's data products are discoverable in the OneLake catalog.

Real-Time Intelligence

Business Events, generally available. Governed business signals published once for both operational and analytical consumers, with a default eventhouse keeping every event. Eventstream and Activator can publish them without code, in preview.
Copilot in Real-Time Intelligence: Real-Time Dashboards built from a description, help with eventhouse administration and schema recommendations. No stage stated.
Event Schema Registry, generally available, for contracts between event producers and consumers.
New Eventstream sources, generally available: Cribl, Solace PubSub+, SAP Datasphere and MongoDB change data capture. A framework for custom streaming connectors is in preview, and Azure Event Hubs sources can sign in with workspace identity, also in preview.
Activator gained stateful conditions and the ability to start Copy jobs, dataflows and functions, all generally available.

SQL

SQLCon ran alongside FabCon, and the SQL news was about where SQL runs, how far it scales and how much of its care an agent can take on.

SQL Server on Azure Local, generally available, both connected and disconnected. The disconnected mode, Azure Local Disconnected Operations, brings a local control plane for sovereign or air-gapped sites. Foundry Local on Azure Local is in preview for AI next to that data.
Hyperscale: 160 and 192 vCore premium-series options, generally available; serverless auto-pause and auto-resume, in preview; elastic pools of up to 50 databases and 128 TB, in preview; and a migration path to Hyperscale from Azure Arc, in preview. Microsoft cited SAP Commerce Cloud, which moved more than 6,000 production databases to Hyperscale.
Vector search. DiskANN vector indexes are generally available in Azure SQL Database, Managed Instance and SQL database in Fabric, with full insert, update and delete support.
Automatic index compaction, generally available in Azure SQL. The platform handles index bloat instead of a maintenance script.
Database Hub in Fabric, in preview. One place to monitor health, assess risk and enforce governance across SQL Server, Azure SQL, PostgreSQL, Cosmos DB and SQL databases in Fabric.
Database agents for SQL and PostgreSQL, preview starting in Database Hub and VS Code. They reason over database signals, find bottlenecks, recommend changes and, within guardrails, help carry them out.
SSMS: GitHub Copilot agent mode, SQL Projects, Schema Compare and SQL Formatter, all generally available. SQL Projects put schema changes in source control.
SQL database in Fabric: vector search and migration from SQL Server, generally available; workspace-level Private Link, in preview.
Managed Instance: flexible memory and higher log rate for Business Critical, generally available, plus several previews, including multiple databases per availability group for the MI link.
For developers: SQL agent skills and a new Azure SQL Dev Hub, in preview, and vector and JSON support in Data API builder.

Platform and billing

On-demand billing extends pay-for-use billing across Fabric workloads, and Fabric zero-provisioned lets a team start with Fabric and OneLake without a capacity commitment. No stage stated.
Capacity overage, generally available and on by default for new F SKU capacities.
Fabric in GCC High, for United States government customers.

Audit and control over agents

This got less stage time than Copilot, and it matters more. An agent that acts needs the same controls a person has, plus a record of what it did.

Policies in Fabric, in preview. Administrators set rules for item creation and workspace security, with external sharing policies coming soon. Policy lifecycle events go to the Microsoft 365 audit log and Purview Audit.
Approval activity in Data Factory. A pipeline can pause for a human decision, with an agent-assigned risk score and an audit trail of who approved what.
Observability in Fabric, in preview, for investigating failures and performance across workloads.
Unified workspace monitoring, in preview, including an agents view in the Monitor Hub.
Sensitivity labels and data loss prevention flow into Copilot, and semantic models can be marked to keep Copilot out of them.
OneLake catalog Govern experience, generally available, plus governance skills for agents.
OneLake Table Read API, in preview, which enforces row and column level security for any engine that reads through it.
OneLake catalog search API, in preview, also available over MCP.
Outbound access protection for OneLake and Eventhouse, generally available: external shortcuts, copies and outbound traffic are blocked by default and allowed only to approved destinations. For maps and the operations agent it is in preview, as is an allowlist of approved external domains.
Data loss prevention can restrict access, generally available. When a Microsoft Purview policy finds sensitive information, access is limited automatically.
Connection allowlists, generally available, limiting cloud connections to trusted tenants and approved sign-in methods.
OneLake security, generally available: views of who holds each role and what data each role can reach, plus view users in role. Granular Spark lineage is coming soon.

Standards

Apache Ossie. Microsoft and Snowflake committed to a converter between Ossie, Snowflake semantic views and Power BI semantic models, and later to DAX as an Ossie query language and ontology support in Ossie.
RDF and OWL in the ontology, listed above.
MCP across Fabric IQ, the Fabric platform, Eventhouse, Data Factory and the OneLake catalog.

Across the industry

Every other major platform built its own version this year:

Databricks made metric views generally available in April, put Genie Ontology into public preview in June, launched Genie One at Data + AI Summit and made the Genie One MCP server generally available the same week as FabCon.
Snowflake renamed Snowflake Intelligence to CoWork, opened Semantic Studio in private preview and exposed semantic views over MCP. It joined Microsoft in the Ossie commitment above.
Google put a managed Looker MCP server and Knowledge Catalog into preview and made conversational analytics in BigQuery generally available.
Oracle added annotations that describe what data means in AI Database 26ai and per-user data security for agents.
AWS made Bedrock AgentCore generally available with MCP and A2A support.
Shared plumbing moved too. The Open Semantic Interchange entered the Apache Incubator as Apache Ossie, and MCP went stateless in its July revision.

What I take from it

Context is now a platform feature. Every vendor has one, and each describes meaning inside its own platform. Most companies run more than one of these platforms, so the same customer or the same revenue figure can now be defined four times, once per platform.

That makes interchange the thing to watch. Ossie, RDF and OWL import and export, and MCP are what let a definition written once reach agents everywhere. I'd judge each vendor on what it supports in production, not on what it announced in Barcelona.

Governance is moving from data to actions. Policies, approval steps, risk scores, audit logs and an agents view in monitoring are the start of treating agents like employees: they get permissions, and they leave a paper trail. I expect this to be the part of FabCon that still matters in two years.

Drafting is getting automated. Deciding is not. An agent can propose what "active customer" means after reading your documents and semantic models. Finance, sales and operations still have to agree on it. The record of who decided, and when, is what lets a business stand behind an agent's answer.