Data, Agents, and the Fabric of Everything
My notes and takeaways from the Microsoft Fabric Conference 2026 — from the keynote to sessions on AI agents, Purview, and responsible AI.
Microsoft IQ: A Unified Intelligence Layer
The keynote set the tone for the entire conference: AI agents should be empowered with the same knowledge and context as your employees — understanding how the business works, how employees work, and drawing on curated, governed data to do so.
Microsoft framed this vision around three pillars of intelligence:
- Work IQ — AI agents embedded in Office 365 applications
- Foundry IQ — curated enterprise knowledge, with visibility into SharePoint and other knowledge sources
- Fabric IQ — the state of your core business data, surfaced through OneLake and Power BI semantic models
One memorable aside from the keynote: Fabric releases happen every single week. If Fabric feels different every time you log in, that's by design.
CI/CD and DevOps in Fabric
End-to-end CI/CD capabilities got a meaningful upgrade. Highlights include selective branching, the ability to compare workspace changes before committing, visualization of git branch structure, parameterized deployments, and a diff view before check-in. A variable library feature now lets you select environment values from a dropdown rather than typing IDs manually — a small but welcome quality-of-life improvement.
OneLake Catalog
The OneLake Catalog gains AI-powered data descriptions and governance tabs for sensitivity labels. Security is now define-once-enforce-everywhere — a significant step for large organizations managing sprawling data estates.
Capacity Management
New capacity tooling includes real-time alerting and usage tracking via the Real-Time Hub, workspace surge protection to keep mission-critical workloads alive, and automatic overage billing controls to prevent surprise throttling.
SQL Gets an AI Makeover
SQL Server 2025 arrives with AI capabilities built in from the ground up, along with JSON support baked directly into the engine. GitHub Copilot is now available inside SSMS 2022, enabling chat-with-your-data experiences, schema-aware grouping, improved export options, and SSDT project support directly in the IDE.
Azure SQL highlights
- SQL MCP (Model Context Protocol) support
- Larger core options for demanding workloads
- Vector index improvements for AI retrieval scenarios
SQL database in Fabric
The SQL database experience inside Fabric now reaches engine parity, is positioned as enterprise-ready, and includes a migration assistant to help teams move existing workloads in.
Database Hub
One of the more exciting announcements: a Database Hub providing a unified view of all databases in your environment — including on-premises databases. From here, database agents can monitor and manage an entire fleet, track capacity utilization, handle compliance and security, and proactively alert on and remediate issues before they become outages.
Unifying the Data Estate with OneLake
A recurring theme of the day was the idea of OneLake as the "OneDrive for data" — a single logical store where data can either be physically stored or referenced via shortcuts. The session walked through several major capabilities:
Shortcuts and mirroring
- Cross-cloud shortcuts let you point to data wherever it lives
- SAP and Oracle mirroring are generally available
- SharePoint mirroring is coming soon
- Shortcut transformations allow in-place views: CSV to table, Excel sheets to individual tables, Parquet and JSON transformations
Mirroring improvements
Delta change data feed support, fine-grained incremental change capture, and — notable for Databricks shops — bidirectional OneLake sharing, meaning you can now read and write both ways between Fabric and Databricks.
Data Factory and Data Warehouse
The Data Factory Migration Assistant now offers full parity with the Fabric Data Factory, removing a major blocker for teams looking to migrate from ADF. New capabilities in the data warehouse include materialized lake views (enabling cross-lakehouse queries), multimodal AI functions, custom SQL pools, AI functions, and real-time alerts. Graph capabilities are also coming to Fabric.
Fabric for planning and analysis
Fabric is expanding into enterprise planning territory — budgets, forecasts, goals, plans vs. actuals. Two agent types got their own spotlight: Data Agents act as virtual analysts that answer questions about your data, while Operation Agents provide 24/7 monitoring and can act on data autonomously. Semantic models remain the key to making data visible and understandable to AI.
Live Pools — pre-warmed compute pools that can be scheduled to spin up at a specific time — are a welcome addition for teams running time-sensitive workloads.
The Agent Revolution: From Single Models to Orchestrated Fleets
This session reframed where we are in the agent evolution. Early AI agents were single-model, manually coded. Next-generation agents are multi-model with automated workflows. The direction is toward dynamic, AI-driven orchestration with end-to-end security — eventually reaching a state where AI creates its own CI/CD pipelines.
Microsoft Azure AI Foundry
Foundry is Microsoft's platform for building, orchestrating, and governing AI agents and applications. Its pillars include:
- AI App and Agent Orchestration
- Model selection and hosting
- Knowledge tools and retrieval
- Observability and agent controls
- Fine-tuning, customization, and edge/local deployment
Foundry Agent Service
Open source and designed for interoperability, Foundry Agent Service supports "hosted agents" so other systems can hook in. Microsoft introduced the Microsoft Agent Framework — an open-source SDK with open standards for building and orchestrating intelligent agents.
Foundry IQ
Built on Azure AI Search, Foundry IQ is specialized for enterprise AI retrieval. It can ingest an AI-optimized copy of your data, index remote sources, and serve as the knowledge backbone for your agents. The OneLake Catalog is also surfaced within Foundry, connecting data governance and AI retrieval in a single plane.
The new Foundry URL is https://ai.azure.com.
Governing Data at Scale: Purview and AI-Driven MDM
Fabric enables teams to work with data closer to the source, but that freedom requires a strong governance layer. Purview is that centralized control plane, spanning:
- Data security
- Data governance
- Risk and compliance
- Data loss prevention
Key capabilities
- Information Protection — sensitivity labels and policies (Public / Internal / Confidential)
- Data Loss Prevention — monitor, detect, and act on sensitive data movement
- Insider Risk Management — protection against both malicious and inadvertent insider risk
- Data Security Posture Management — surface, discover, and assess risks proactively
AI-Driven Master Data Management (MDM) on Fabric
Perhaps the most impactful announcement in this session: AI-driven MDM built directly on Fabric, designed to create the "Golden Record" — a single, trusted source of truth for key business entities. Without MDM, organizations suffer from duplicates across systems, inconsistent formats and attributes, and limited cross-system visibility. AI-driven MDM addresses this by:
- Cleaning data using AI-powered rule suggestions
- Enriching records by filling in missing data
- Standardizing values (e.g., "US" vs. "USA")
- Consolidating from multiple source systems
Practical AI: Models, MCPs, and Responsible Deployment
This session — featuring Bob Ward, whose open-source presentations are available at https://aka.ms/bobwardms — was a grounding counterbalance to the big-vision keynote energy. Key takeaways:
The architecture of AI
- AI models are algorithms — your AI apps control everything
- Smaller models are cheaper and often sufficient; bigger is not always better
- MCP (Model Context Protocol) is the standard method to run and discover tools — described as "COM objects for AI" or "USB for AI"
- Agents use tools and/or MCP servers to act on your behalf
Tools mentioned
- LM Studio — for experimenting with local models
- Azure AI Gateway (AIPM) — can protect against jailbreaking by scanning prompts and responses
- GitHub Copilot Instructions — use as a system prompt to guide AI behavior in your workflows
Responsible AI principles from the session
- Use vetted and trusted AI models; audit and govern AI model usage
- Tune the temperature for responses appropriate to your use case
- Better prompts = better responses; evaluate the quality of your RAG pipeline
- Always consider what data permissions are required
- Ask models questions — don't tell them what to do — and do not turn "auto approve" on
- Create a responsible AI policy for your organization
 |
| Bob Ward signing his new book. The photographer didn't wait for me to finish turning before she snapped the picture, so it's blurry. |
That wraps Part 1 of my report. It was a lot to absorb — Fabric's surface area keeps expanding, and the AI agent story is moving fast. More session recaps to follow.
Mark your calendar: the next FabCon is in Atlanta, March 8–12, 2027.
Full agenda: https://aka.ms/FabCon-Agenda
