Microsoft's Fabric IQ and Foundry IQ: Giving AI Agents the Right Data and Context
At Ignite in San Francisco, Microsoft previewed two capabilities aimed squarely at a problem most AI projects hit fast: agents need consistent context and the right data at the right time. Fabric IQ extends the semantic layer from Power BI across Fabric, while Foundry IQ automates retrieval pipelines to feed agents relevant information.
The goal is straightforward. Unify meaning (semantic models) and automate retrieval (RAG) so agents can act without hand-holding. Helpful for teams building real applications-not just demos.
What Fabric IQ Does
Fabric IQ extends Power BI's semantic modeling so the same business entities and relationships apply across operational systems in OneLake. Model it once with governance, and both analysts and AI agents can consume it consistently.
Think: customers, products, policies, accounts-standardized definitions mapped to their sources. That consistency reduces misfires, brittle prompts, and duplicated logic across teams.
- Semantic models live in OneLake and can reflect all connected data sources
- Security and compliance run through Fabric governance
- AI agents can query data with business context instead of raw tables
As one analyst put it, "The context of these semantic models is critical to successful AI and BI implementations, particularly agentic AI."
What Foundry IQ Does
Foundry IQ, built on Azure AI Search, automates retrieval-augmented generation (RAG) pipelines to deliver relevant data to AI agents. It's integrated with Microsoft Purview for governance.
- Automated retrieval from multiple sources
- Governed data access with lineage and policies
- Feeds agents facts, docs, and context on demand
Put simply: Fabric IQ sets the language; Foundry IQ delivers the content.
Who Benefits Most
- Organizations already invested in Microsoft Fabric, Power BI, Azure AI Search, and Purview
- Teams standardizing data models and enforcing consistent definitions across domains
- Agentic AI initiatives that need governed, production-grade context and retrieval
As one industry voice noted, these features reduce friction more than they add radical novelty-especially for customers already deep in the Microsoft stack. Teams with highly tuned, domain-specific retrieval systems may find Foundry IQ more general than they prefer.
Databases: HorizonDB, SQL Server Updates, and GA Releases
Beyond agents, Microsoft introduced Azure HorizonDB, a PostgreSQL database service aimed at modern application and agent development. PostgreSQL continues to be a go-to for flexible, heterogeneous data work in AI pipelines.
Microsoft also updated SQL Server and made Azure DocumentDB and Fabric databases (a combo of Cosmos DB and a SQL database) generally available. For many enterprises, GA status is the line between pilot and production, so these releases matter.
Why Semantic Modeling Matters Now
Agent performance depends on consistent, high-quality data. Semantic modeling enforces clarity-what entities mean, how they relate, and which metrics count. A vendor group (including Snowflake and Salesforce) is working on an open standard. Microsoft isn't part of it yet, and analysts suggest collaboration on a common semantic model would benefit everyone.
How to Evaluate Fabric IQ and Foundry IQ
- Inventory your current semantic models in Power BI/Fabric; identify gaps across operational systems
- Map key agent use cases to specific entities, policies, and retrieval needs
- Assess retrieval complexity: do you need general-purpose RAG, or domain-specific flows?
- Run a pilot with 1-2 high-impact agent tasks that cross multiple data sources
- Measure: task success rate, latency, hallucination rate, policy violations caught by Purview
Architecture Pattern to Consider
- OneLake as the unified data plane
- Fabric IQ semantic models as the contract for meaning and metrics
- Purview policies for access, PII handling, retention
- Foundry IQ for retrieval (docs, structured data, embeddings) via Azure AI Search
- Agents (Azure OpenAI or others) consuming both semantic context and retrieved facts
Caveats and Limitations
- Both Fabric IQ and Foundry IQ are in preview-production plans should account for change
- Best fit is Microsoft-centric environments; heterogeneous stacks may need adapters or custom RAG
- Teams with bespoke, domain-optimized retrieval may see Foundry IQ as less flexible
Security and Governance
- Use Purview to centralize access policies and lineage; test with least privilege
- Apply row-level and object-level security within semantic models before exposing to agents
- Log agent access paths and decisions for auditability and incident response
Action Plan for IT and Product Teams
- Stand up a Fabric workspace tied to OneLake; define a minimal semantic model for a single product or line of business
- Integrate Purview policies and data classifications early
- Prototype Foundry IQ on a focused agent workflow (e.g., customer entitlement checks + policy retrieval)
- Instrument metrics: grounding coverage, response accuracy, and time-to-decision
- Iterate on entity definitions and retrieval scopes; lock down PII handling
Competitive Context
Analysts see AWS, Google Cloud, Microsoft, and Oracle clustered near the top for database and AI capabilities, with a slight edge today to Oracle and Google in some areas. Microsoft's moves are pragmatic: strengthen semantics, simplify retrieval, and push GA status where customers need production readiness.
Where to Learn More
Skills and Team Readiness
If your roadmap includes agentic workflows tied to governed data, upskilling on semantics, RAG, and policy enforcement will pay off. A practical route is role-aligned learning paths that connect data modeling with applied AI.
Bottom line: Fabric IQ and Foundry IQ make it easier for Microsoft shops to move from siloed data and brittle prompts to agents that carry real context. Add HorizonDB and GA database updates, and you have a firmer path from prototype to production-without reinventing your data stack.
Your membership also unlocks:
