Foundry IQ + Fabric IQ: The New Semantic Superpowers Behind Microsoft's AI Strategy
If you've ever watched an AI agent hallucinate its way through a business question because it couldn't tell the difference between your Q3 sales deck and someone's leftover pizza order in Teams, congratulations—you've experienced the joy of fragmented enterprise context. Microsoft just announced they're done with that nonsense.
At Ignite 2025, Microsoft unveiled what might be the most architecturally ambitious piece of their agent-first strategy yet: a universal context layer that unifies intelligence across Work IQ, Fabric IQ, and Foundry IQ. Think of it as building "one brain to rule all enterprise data"—a shared semantic foundation that finally lets AI agents understand what you're doing, what your business data actually means, and where to find the information they need without making stuff up.
This isn't just another feature drop. This is Microsoft rewiring the connective tissue between productivity apps, analytics platforms, and AI development environments to create something that's been missing from the enterprise AI conversation: cross-system semantic reasoning that doesn't make you want to throw your laptop out a window.
The Unified Context Layer: One Brain for All Your Data Silos
Here's the setup. For years, enterprise data has lived in three separate universes:
Collaboration data (emails, chats, documents) lived in Microsoft 365
Analytics data (metrics, KPIs, semantic models) lived in Power BI and Fabric
Custom application data scattered across Azure services, on-prem systems, and whatever your development team cobbled together last quarter
Each had its own understanding of what "customer," "revenue," or "project" meant. Each spoke its own semantic language. And AI agents? They were left playing telephone between these systems, trying to piece together context like a detective with half the case files missing.
Microsoft's new unified context layer changes that by introducing three interconnected intelligence systems:
Work IQ: The Productivity Brain
Work IQ is the intelligence layer that powers Microsoft 365 Copilot and agents. Built on data, memory, and inference, it understands your work inside and out. It connects to organizational and personal data—files, emails, meetings, SharePoint libraries—while building memory from your preferences, habits, and workflows.
With the new conversational memory feature (available through Microsoft's Frontier early access program), Work IQ can now retain context and specific details across sessions. This means Copilot doesn't treat every conversation like meeting you for the first time. It knows your custom instructions, your saved preferences, and the context from previous interactions.
And here's where things get interesting for those of us who've battled SharePoint metadata: Work IQ now supports reasoning over structured metadata in SharePoint document libraries. Finally, an AI that understands your SharePoint library better than your entire marketing department. When you ground a Copilot prompt on a SharePoint library of vehicle spec sheets with metadata like make, model, and engine size, it delivers answers that are actually accurate, complete, and relevant—not just keyword soup.
Fabric IQ: The Business Data Brain
Now in preview, Fabric IQ extends the unified semantic layer that Power BI has been providing across 20 million models to business operations. This is huge.
Fabric IQ unifies all your data—whether it lives in OneLake, on premises, hybrid, or multicloud environments—under a semantic model of business entities and their relationships. You define what a "customer" or "revenue transaction" means once, and that definition becomes the source of truth everywhere: across analytics, apps, and AI agents.
This gives teams and agents a live, connected view of the enterprise. No more "wait, which customer table are we using?" or "is this the current revenue figure or last quarter's?" The semantic model handles that, preserving governance and lineage while improving reasoning quality.
Since all the data resides in OneLake (either natively or through shortcuts and mirroring), you get these benefits regardless of where your data actually lives. Microsoft has essentially turned your data platform into a semantic layer that speaks one consistent language.
Foundry IQ: The RAG Brain
Also in preview, Foundry IQ is the next generation of retrieval-augmented generation (RAG), and it's where things get properly technical in the best way possible.
Foundry IQ is a fully managed knowledge system designed to ground AI agents. Built on Azure AI Search, it lets agents connect to one knowledge base running a knowledge retrieval engine over multiple data sources—indexed or federated—including Azure data services, Microsoft 365 SharePoint, Fabric IQ, and the web.
But here's what makes it special: Foundry IQ automates RAG data pipelines for multimodal data and performs agentic retrieval—query planning, iterative search, reflection, and synthesis—across knowledge sources to maximize context while respecting user permissions. Integration with Microsoft Purview ensures compliance, data security, and governance throughout.
This is Microsoft saying, "We're done with basic vector search and hoping for the best." Foundry IQ brings intelligence to the retrieval process itself.
Next-Gen RAG: When Your AI Actually Finds the Right Information
Let's talk about what makes this unified context layer such a game-changer for RAG pipelines.
Traditional RAG workflows suffer from a few predictable problems:
Semantic drift: The same business concept means different things across systems
Context fragmentation: Relevant information is scattered across data silos
Naive retrieval: Vector search alone can't reason about what information is actually needed
Permission chaos: Great, your agent found the answer—in a document the user can't access
The unified IQ layer addresses all of this.
With Fabric IQ providing the semantic backbone, your agents understand business concepts consistently. When an agent needs information about "Q3 sales performance," it knows exactly what entities, relationships, and metrics that includes—and where to find them.
Foundry IQ takes that semantic understanding and performs sophisticated agentic retrieval. Instead of just doing a simple vector similarity search, it:
Plans the query based on understanding what information is actually needed
Iteratively searches across multiple knowledge sources
Reflects on the results to determine if they're sufficient or if more searching is needed
Synthesizes the findings while respecting permissions
The result? Agents that give you answers grounded in actual enterprise context, not hallucinated corporate fanfiction.
The Enterprise Semantic Brain: Cross-App Intelligence That Actually Works
Here's where the magic happens: When you unify semantics across Work IQ, Fabric IQ, and Foundry IQ, agents can finally reason over data that once lived in completely separate universes.
Real-world scenario: A sales operations agent needs to answer, "How is our pipeline trending against last year's Q4, and are there any delays flagged in customer conversations?"
Without the unified context layer, this requires:
Accessing your CRM or Fabric analytics for pipeline data
Searching through Teams and email for customer conversations
Manually correlating "delays" mentioned in conversations with pipeline stages
Hoping nothing gets lost in translation between systems
With the unified IQ layer:
The agent understands "pipeline" as a business entity with consistent semantic meaning across Fabric IQ
Work IQ provides access to Teams and email conversations with proper context
Foundry IQ performs agentic retrieval across both domains, understanding which information is relevant and how it connects
The agent synthesizes an answer that combines quantitative pipeline data with qualitative signals from customer conversations
All while respecting permissions and maintaining governance through Microsoft Purview.
Another example: An agent in Word or PowerPoint (both now integrated with Work IQ) can automatically select relevant sources from SharePoint, Fabric OneLake, and Teams conversations to generate content that's actually grounded in your organization's knowledge—not generic corporate templates that could apply to any company.
This is what "one brain to rule all enterprise data" looks like in practice.
What This Means for Hallucination Reduction
Let's address the elephant in the room: hallucinations.
The primary cause of AI hallucinations isn't model failure—it's context failure. When agents don't have proper context, or when they can't reconcile conflicting semantic definitions, they fill in the gaps with plausible-sounding nonsense.
The unified IQ layer attacks this at the architectural level:
Semantic consistency (via Fabric IQ) means agents aren't trying to reconcile three different definitions of the same business concept
Agentic retrieval (via Foundry IQ) means agents can validate their understanding through iterative search and reflection before committing to an answer
Context retention (via Work IQ's conversational memory) means agents maintain understanding across sessions instead of starting from zero every time
Does this eliminate hallucinations entirely? No. But it dramatically reduces the surface area where they can occur by giving agents the semantic grounding and retrieval sophistication they need to find truth instead of inventing it.
The Integration Stack: Where This All Lives
For the Azure and data platform practitioners in the room, here's how the pieces fit together technically:
Work IQ sits at the Microsoft 365 layer, integrating with Copilot, SharePoint, and productivity apps
Fabric IQ lives in Microsoft Fabric, extending the semantic layer across OneLake and connecting to operational systems through shortcuts and mirroring
Foundry IQ operates within Microsoft Foundry (the developer platform) and integrates with Azure AI Search, Azure data services, and Microsoft Purview
Agent 365 provides the management and governance layer across all three, giving IT visibility and control over the entire agent ecosystem
This isn't a bolt-on feature. This is Microsoft rebuilding the semantic backbone of their entire AI strategy.
What This Means for IT Pros
If you're responsible for deploying AI agents in your organization, here's what you need to know:
Start thinking semantic-first. The unified IQ layer only works if you invest in semantic modeling. That means:
Defining business entities and their relationships clearly in Fabric IQ
Populating SharePoint with structured metadata (manually or with Knowledge Agent)
Taking advantage of Work IQ's conversational memory to build agent context over time
Plan for integration. These features are in preview (Fabric IQ and Foundry IQ) or rolling out through early access programs (Work IQ conversational memory). Start testing with the Frontier program if you have access. Understand how your existing Azure AI Search, Fabric, and Microsoft 365 implementations will integrate with these new IQ layers.
Governance matters more than ever. With agents having access to unified context across your entire enterprise, Microsoft Purview integration becomes critical. Make sure your data classification, sensitivity labels, and permission structures are tight before you turn agents loose on your semantic brain.
RAG pipelines are evolving. If you've built custom RAG solutions using Azure AI Search, understand how Foundry IQ's agentic retrieval capabilities might augment or replace parts of your pipeline. The shift from naive vector search to agentic retrieval is significant.
This is foundational architecture. The unified context layer isn't a feature you enable—it's the substrate on which Microsoft's entire agent-first strategy runs. Every agent announcement at Ignite builds on this foundation. Understanding it helps you understand where Microsoft is taking AI development.
The Bottom Line
Microsoft is betting that the future of enterprise AI hinges on solving the context problem. Not just retrieving documents. Not just matching vectors. But creating a unified semantic understanding that spans collaboration, analytics, and application data—and making that understanding accessible to AI agents in a governed, permissioned, and architecturally sound way.
Work IQ, Fabric IQ, and Foundry IQ aren't three separate products. They're three facets of the same architectural vision: one brain to rule all enterprise data.
If Microsoft pulls this off—and the technical foundations look solid—they're not just improving RAG pipelines. They're redefining what it means for AI to understand your business.
And personally? I'm just thrilled we might finally have agents that understand SharePoint metadata. It only took a universal context layer and a complete rethinking of enterprise semantics, but we got there.
