How Microsoft Fabric IQ and Foundry IQ Are Revolutionizing AI Agents

At Microsoft Ignite 2025, Microsoft unveiled a fundamental shift in enterprise AI architecture. Two complementary technologies Fabric IQ and Foundry IQ represent a new approach to making AI agents truly understand business context, not just process data.

This isn’t an incremental improvement. It’s a reimagining of how enterprise AI systems think, reason, and act.

The Problem: AI Without Context Is Just Guessing

For years, enterprise AI has followed the same playbook: collect more data, add more tools, build bigger models. But there’s been a persistent gap between where data lives and how AI systems reason about it.

AI agents have struggled with fragmented definitions, conflicting data interpretations, and the inability to understand what business concepts actually mean. Ask an AI about “customer churn” and it might pull from three different systems with three different definitions producing answers that are technically accurate but practically useless.

This is the challenge that Fabric IQ and Foundry IQ were built to solve.

Fabric IQ: The Semantic Foundation

Fabric IQ is a workload within Microsoft Fabric that unifies data across OneLake, including lakehouses, eventhouses, and semantic models, and organizes it according to the language of your business. It transforms Microsoft Fabric from a unified data platform into a unified intelligence platform.

The data isn’t just stored; it’s organized and exposed to analytics, AI agents, and applications with consistent semantic meaning and context.

The Core: Ontology

At the heart of Fabric IQ lies the Ontology item, a machine-understandable vocabulary of your business. Think of it as the authoritative dictionary that defines:

• Entity Types: Reusable logical models of real-world business concepts like Customer, Product, Shipment, or Sensor
• Properties: The facts and attributes associated with each entity type
• Relationships: How entity types connect to each other
• Constraints and Rules: Business logic that keeps representations consistent

When you define “Customer” in the ontology, every tool, dashboard, notebook, and AI agent in your organization uses that same definition. No more conflicting interpretations.

Data Binding: Connecting Concepts to Reality

The ontology isn’t just theoretical, it connects to real data through a process called binding. Bindings map your ontology definitions to concrete data living in OneLake, including lakehouse tables, eventhouse streams, and semantic models.

This binding layer describes data types, identity keys, column-to-property mappings, and relationship keys across multiple data sources. It enables schema evolution rules, data quality checks, and provenance tracking at the concept layer, turning raw rows and events into governed business objects.

Graph Capabilities

Fabric IQ automatically builds a navigable graph from your ontology. In this graph, nodes represent entity instances and edges represent relationships with metadata attributes. This enables:

• Visual exploration of business context
• Graph algorithm execution (paths, centrality, communities)
• Rule-driven inferences
• Cross-domain reasoning

Want to trace a problem from an order through a shipment to a temperature sensor to a cold chain breach? The graph makes that relationship explicit, queryable, and governed.

Key Components Within Fabric IQ

Fabric IQ includes several integrated items:

• Ontology (preview): The enterprise vocabulary and semantic layer
• Fabric Data Agent: Build conversational Q&A systems using generative AI
• Graph in Microsoft Fabric: Native graph storage and compute for connected data
• Operations Agent: AI agents that monitor real-time data and recommend actions
• Power BI Semantic Model: Curated analytics models optimized for reporting

The Operations Agent

The Operations Agent continuously monitors your business in real time, reasons over live conditions, evaluates trade-offs, and automatically takes actions to advance business outcomes. It uses ontology definitions, rules, and actions to build its playbook for responding to operational situations.

When business constraints are violated, whether it’s a temperature threshold exceeded or inventory dropping below safety stock, the Operations Agent detects it and triggers the appropriate workflows or alerts.

Foundry IQ: Intelligent Knowledge Retrieval

Built on Azure AI Search, Foundry IQ is a unified knowledge layer for agents within Microsoft Foundry. It’s designed to improve response performance, automate RAG (Retrieval-Augmented Generation) workflows, and enable enterprise-ready grounding.

While Fabric IQ provides the semantic foundation for structured business data, Foundry IQ handles the retrieval and synthesis of unstructured knowledge, documents, policies, SharePoint content, and web sources.

Knowledge Bases: Reusable Intelligence

Instead of wiring retrieval logic into every agent, you define a reusable knowledge base around a topic, employee policies, product documentation, support content, and create it in the Foundry portal. Multiple agents and applications can then connect to and be grounded by that same knowledge base.

The benefits are significant:

• Update a knowledge base independently without modifying agents
• Multiple agents share the same knowledge base, avoiding duplicate configurations
• Developers don’t need to manage routing or implement different retrieval strategies per source

Multi-Source Integration

Foundry IQ accesses data across both indexed and remote knowledge sources:

• Microsoft 365 SharePoint
• Fabric IQ
• OneLake
• Azure Blob Storage
• Azure AI Search
• The web
• MCP (Model Context Protocol) sources

For indexed sources, Foundry IQ automatically manages the full indexing pipeline: content is ingested, chunked, vectorized, and prepared for hybrid retrieval.

The Agentic Retrieval Engine

This is where Foundry IQ fundamentally differs from traditional RAG approaches.

When an agent invokes a knowledge base, the system doesn’t just do a simple vector similarity search. Instead, it orchestrates a sophisticated multi-step process:

1. Query Planning: An LLM analyzes the query (including chat history) to identify the underlying information need
2. Query Decomposition: Complex queries are broken into focused subqueries for better coverage
3. Parallel Processing: Subqueries run simultaneously using keyword, vector, or hybrid techniques
4. Semantic Reranking: Results are scored and ranked to identify the most relevant matches
5. Reflective Search: The system reviews retrieved results and issues follow-up queries if needed
6. Synthesis: Results are unified into a response with source references

Reflective Search: Self-Improving Retrieval

Foundry IQ uses a newly trained small language model (SLM) called the semantic classifier to determine if another round of search would improve results for hard queries, while providing a low-latency exit path for easier ones.

The result? An average of 20 points (36%) improvement in end-to-end RAG answer quality compared to brute-force searching all sources at once.

Enterprise Security Built In

Foundry IQ respects user permissions for configured knowledge sources. For SharePoint content, data classifications and sensitivity labels from Microsoft Purview are honored throughout the indexing and retrieval pipeline.

Classified content remains tagged and governed as it flows into knowledge bases, and Purview policies continue to apply when agents are grounded on that data. This closes one of the biggest gaps in DIY RAG implementations, where security rules often have to be approximated or duplicated in application code.

The Power of Integration: How They Work Together

Fabric IQ and Foundry IQ aren’t competing technologies; they’re complementary layers of a unified intelligence architecture.

The Flow

1. Fabric IQ creates the semantic foundation, defining what business concepts mean and how they relate
2. Foundry IQ provides the intelligent retrieval layer that agents use to access grounded knowledge
3. Agents inherit live business context from Fabric IQ through Foundry IQ

Cross-Domain Reasoning

Together, these technologies enable what Microsoft calls “cross-system semantic reasoning.” Consider a scenario where 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?”

• Fabric IQ provides the hard data on pipeline numbers and year-over-year comparisons
• Work IQ (Microsoft’s intelligence layer for M365) surfaces qualitative context from client emails and Teams chats
• Foundry IQ orchestrates the retrieval and synthesis, respecting security boundaries at every step

The result is insights that are both quantitatively accurate and qualitatively rich—dramatically reducing the hallucinations that occur when AI attempts to guess context it doesn’t possess.

Building Agents with Shared Understanding

Developers building agents now have a significant advantage: instead of starting from scratch, agents inherit live business context. Every agent understands the business, not merely has access to the data. They reason using the same model and definitions as human teams.

Multiple agents and teams can collaborate on a shared foundation, exchanging context instead of pulling in conflicting directions. A developer building a new logistics agent doesn’t need to redefine what “shipment,” “route,” or “delivery window” means, those definitions already exist in the ontology.

---

Comparison: Fabric IQ vs. Foundry IQ