Enter your search text in the box above

Select a result to preview

Making AI Agents Smarter with Fabric IQ and Foundry IQ

#fabcon2026 #session #learning #microsoft #fabcon2026-session

Description

Connecting AI agents to company data across SharePoint, databases, and documents is complex. Foundry IQ + Fabric IQ provide secure, unified access through one API. Learn to connect agents to multiple sources, configure permissions, use Agent Factory for deployment, and ensure agents understand business context, not just raw data.

Key Takeaways

My Notes

Action Items

Slides

📥 Download Slides

Making AI Agents
Smarter with
Fabric IQ & Foundry IQ
Treb Gatte | Marquee Insights
Microsoft Data Platform / AI Platform MVP
Session Agenda
3 Demos | 300-Level Technical Depth
0-10 min
The Problem & Microsoft IQ Intelligence Layer
10-20 min
Fabric IQ Deep Dive: Ontology, Graph, Agents
20-30 min
DEMO 1: Build Ontology from Power BI + Eventhouse
30-42 min
Foundry IQ: Agentic Retrieval & Python SDK
42-48 min
DEMO 2: Knowledge Base + Agent Framework Code
48-55 min
DEMO 3: End-to-End Agent + Ontology + KB
55-60 min
Agent Factory, Architecture, Resources, Q&A
Why AI Agents Fail Today
Agents don't fail for lack of data — they fail for lack of meaning.
Fragmented Data
Every team rebuilds RAG pipelines, chunking logic, and
embeddings from scratch
Security Gaps
Homegrown pipelines lack consistent access control
and governance
No Business Context
Agents see tables and columns, not business entities,
relationships, and rules
Semantic Drift
Each department defines 'Customer' or 'Revenue'
differently — agents inherit the confusion
Microsoft IQ Intelligence Layer
Universal context for every agent — from productivity signals to business data to enterprise knowledge
Work IQ
Microsoft 365
Fabric IQ
Microsoft Fabric
Foundry IQ
Microsoft Foundry
Organizational signals from emails,
chats, documents, meetings, and
workflows.
Ontology + Graph + Semantic Models
over OneLake. Defines business
entities, relationships, rules, and
actions.
Knowledge bases with agentic retrieval
built on Azure AI Search. Single API for
multi-source RAG.
Powers M365 Copilot with user-specific
and org-specific context through an AI
feedback loop.
Turns data into operational intelligence
that agents can reason over.
Automatic indexing, vectorization, query
planning, and permission enforcement.
Fabric IQ: Data Platform to Intelligence Platform
Fabric IQ is a workload in Microsoft Fabric that organizes data in OneLake according to the language of
your business, then exposes it to analytics, AI agents, and applications with consistent semantic
meaning.
Ontology (Preview)
Define entity types, properties, relationships, rules, and bind them to real
OneLake data
Graph (Preview)
Native labeled property graph with GQL support for multi-hop reasoning and
impact analysis
Data Agent
Natural language Q&A grounded in ontology — agents understand business
terms, not raw columns
Operations Agent
Monitors real-time data, evaluates business rules, and triggers autonomous
actions
Power BI Semantic Model
Bootstrap ontologies from existing BI models — your investment carries
forward
Ontology Deep Dive: Core Concepts
Entity Types
Properties
Relationships
Reusable model of a real-world concept (Shipment,
Product, Sensor) — standardizes naming across teams
Key Benefits
No-code visual tools for business
experts
Named facts with declared data types, semantic
annotations, and quality checks
Bootstrap from existing Power BI models
Governed: IT secures, versions, and
approves
Typed, directional links between entities with
cardinality rules (Customer → Orders)
AI-ready grounding for all agent types
Data Bindings
Rules & Constraints
Connect definitions to lakehouse tables, eventhouse
streams, and Power BI semantic models
Cross-domain reasoning via graph
traversal
No additional licensing fees
Business rules that agents evaluate — turning tacit
knowledge into automated operations
Operations Agent uses these as playbook triggers for autonomous action.
Ontology Technical: Data Binding & Graph Queries
Data Binding Architecture
GQL Graph Query Examples
Supported Sources (See prerequisites):
Cold-chain risk analysis:
OneLake tables
Order > Route > Asset > Sensor
Eventhouse streams (KQL)
Influence path discovery:
Customer > Household > Product
Binding Process:
Predictive maintenance:

  1. Define entity type + properties
    Well > Compressor > Sensor > WorkOrder
  2. Select source table/stream
  3. Map columns to properties
  4. Set identity keys for instances
  5. Configure relationship keys
  6. Refresh graph model
    Fraud detection:
    Customer > 3x transaction increase within
    2 hops of prior fraud alerts
    Key: GQL for graph, KQL for streams.
    Ontology orchestrates federated plans.
    Predicates push down to native engines.
    DEMO 1
    Build an Ontology from
    Power BI Semantic Model
  7. Generate ontology from RetailSalesModel — auto-creates Store, Product, SaleEvent
  8. Walk through entity types, properties, and auto-mapped relationships
  9. Add Freezer entity type — bind FreezerTelemetry from Eventhouse KQL
  10. Define cross-source relationship: Store operates Freezer
  11. Preview entity instances — show time-series freezer data through ontology
  12. Query graph: 'Show all freezers in Paris store' (StoreId = S-PAR-01)
  13. Data Agent NL query: 'Top product by revenue across all stores?'
    Demo 1: Technical Details & What to Watch For
    Semantic Model →
    Ontology Mapping
    Cross-Engine Data
    Binding (Key Technical
    Differentiator)
    • Tables become
    entity types.
    • Columns become
    properties.
    • DAX relationships
    become ontology
    relationships with
    cardinality.
    • Measures are NOT
    imported —
    ontology focuses on
    structural
    semantics.
    • Freezer entity pulls
    identity from
    lakehouse (static
    metadata) and
    time-series from
    Eventhouse (KQL).
    • Data binding:
    “Integrate without
    copying source
    data” + requirement
    that data be
    prepared and in
    Fabric
    Graph Model Refresh
    Query Pushdown
    • Schema changes
    trigger automatic
    reingestion (Preview
    limitation) but the
    instance data
    updates requires a
    manual or
    scheduled refresh,
    once bindings are
    configured.
    • Ontology
    experience uses
    Graph (GQL) and
    KQL/Eventhouse
    • Graph query builder
  1. Open Foundry portal → Create project → Build → Create Knowledge Base
  2. Add knowledge sources: SharePoint document library + Azure Blob Storage
  3. Configure retrieval reasoning effort (minimal /low/ medium)
  4. Test in playground: 'What is our return policy for damaged items?'
  5. Show query diagnostics: query plan, sub-queries, source routing
  6. Python SDK: ~20 lines — AIProjectClient, create_agent, threads
  7. MCP endpoint: {search_endpoint}/knowledgebases/{name}/mcp
    Demo 2: Python SDK & MCP Configuration
    Azure AI Search Context Provider (~20 lines of Python):
    from azure.ai.projects import AIProjectClient
    from azure.identity import DefaultAzureCredential
    client = AIProjectClient(endpoint, DefaultAzureCredential())
    agent = client.agents.create_agent(model='gpt-4.1-mini', tools=[kb_tool])
    thread = client.agents.threads.create()
    run = client.agents.runs.create_and_process(thread_id, agent_id)
    MCP Tool Configuration:
    server_url: '{endpoint}/knowledgebases/{name}/mcp?api-version=2025-11-01-preview'
    allowed_tools: ['knowledge_base_retrieve']
    require_approval: 'never'
    Note: Per-user auth uses Azure OpenAI Responses API (not MCP headers in Preview).
    Prerequisites: Azure AI Search (Standard+), Foundry project, Python 3.10+
    Connecting Agents to Multiple Data Sources
    Fabric IQ — Ontology Sources
    Foundry IQ — Knowledge Sources
    (Structured / Operational Data)
    (Unstructured / Document Data)
    ✓ Lakehouse Tables (Delta Lake)
    ✓ Eventhouse Streams (KQL)
    Indexed (auto-vectorized):
    ✓ Microsoft SharePoint
    ✓ Azure Blob Storage
    ✓ Fabric OneLake files
    Federated (queried remotely):
    ✓ Public Web content
    ✓ MCP Servers (Private Preview)
    Ontology unifies all under entities.
    No additional Fabric SKU, but consumption applies
    ✓ Work IQ signals (M365)
    One KB, one API, many agents.
    Security, Permissions & Governance
    Microsoft Entra ID:
    End-to-End Identity
    Enforcement
    Queries execute under the caller's Entra identity. Agents return only authorized content.
    Managed identities for service-to-service. Search Index Data Reader role on search
    service.
    Microsoft Purview:
    Sensitivity Labels
    Labels travel with documents from source through indexing to query results. Foundry IQ
    respects labels at query time. Classification and compliance without custom code.
    Document-Level ACLs
    Access control lists synchronized from supported sources. Permissions enforced perdocument, not per-index. Different users, same KB, different results.
    Ontology Governance
    (Fabric IQ)
    Version definitions. Validate schema changes before publishing. Monitor health via
    Fabric monitoring tools. IT secures and approves; business enriches meaning.
    Security rules enforced at the data layer regardless of which engine (Spark, SQL, KQL,
    OneLake Security (Preview) Power BI) accesses the data.
    DEMO 3
    End-to-End Agent with
    Ontology + Knowledge Base
  8. Create Data Agent in Fabric → connect to ontology → add custom instructions
  9. NL query: 'Which stores had freezer temperature breaches?' → graph traversal
  10. Create Foundry Agent → connect to knowledge base → MCP tool
  11. Show agentic retrieval: multi-step reasoning with extractive citations
  12. Permission-aware: User A (store mgr) vs. User B (director) → different results
  13. Same agent, same KB, same ontology — Entra ID enforces at query time
  14. Deployment: M365 via Agent 365 integration
    Agent Integration Patterns
    Pattern 1: Fabric
    Data Agent +
    Ontology
    Standalone Fabric item. Custom instructions + example queries.
    Pattern 2: Foundry
    Agent + Knowledge
    Base (MCP)
    tools=[{type:'mcp', server_url:'{endpoint}/knowledgebases/{name}/mcp', allowed_tools:['knowledge_base_retrieve']}]
    Pattern 3: MultiAgent with Microsoft
    Agent Framework
    Orchestrate specialist agents: one grounded in ontology (structured), one in KB (documents), one for actions (Power Automate).
    Unified AutoGen + Semantic Kernel.
    Pattern 4: Hosted
    Agents in Foundry
    Deploy custom-code agents to managed runtime.
    Deploy to Teams, Copilot Studio, or custom apps.
    Unlike Copilot, Data Agents have personalized context.
    Agent instructions: 'Always provide annotations using MCP tool and render as citations.'
    No K8s.
    Bring Agent Framework, LangGraph, or CrewAI.
    Auto-scaling, monitoring, identity built in.
    Operations Agent: From Insights to Autonomous Action
    Operations Agents continuously monitor, reason, and act in real time.
    Monitor
    Detect
    Evaluate
    Act
    • Real-time
    Eventhouse
    streams bound in
    ontology
    • When business
    constraints are
    violated (ontology
    rules)
    • Trade-offs using
    relationship
    context (graph
    traversal)
    • Trigger workflows,
    alerts, updates
    (Power Automate,
    custom actions)
    Supply Chain Example:
    Ontology models: Routes, Trucks, Warehouses, Sensors, Deliveries.
    Rule: If congestion data shows route delay > 30 min → Reroute trucks, notify warehouse, update delivery ETAs.
    Context: Graph traversal identifies which deliveries are affected downstream.
    Data Agent = reactive Q&A. Operations Agent = proactive autonomous action.
    Both share the same ontology grounding for consistent business understanding.
    Microsoft Agent Factory
    A single program to move from experimentation to execution — build, deploy, and scale AI agents
    Agent Pre-Purchase
    Plan (P3)
    One metered plan with Agent Commit
    Units (ACUs).
    Use flexibly across Copilot Studio +
    Microsoft Foundry.
    1-year commitment. No upfront
    licensing. Simplifies procurement for
    multi-agent deployments.
    Forward Deployed
    Engineers (FDEs)
    Microsoft AI engineers embedded with
    your team.
    Deep expertise in IQ technologies.
    Co-innovate on production-grade
    solutions. Rapid prototyping to
    deployment.
    AI Fluency
    Training
    Tailored learning plans aligned to
    business goals.
    Live instructor-led sessions and
    webinars.
    Role-based: executives, developers,
    data professionals. Upskill the entire
    workforce.
    Architecture: Putting It All Together
    USER SURFACES
    Microsoft 365 | Teams | Custom Apps | Power Apps | Web
    AGENT RUNTIME
    Foundry Agent Service | Microsoft Agent Framework | Copilot Studio | Agent 365
    FOUNDRY IQ — KNOWLEDGE
    Knowledge Bases → Agentic Retrieval → Entra ID ACLs → Purview Labels → MCP
    FABRIC IQ — SEMANTIC
    Ontology → Graph (GQL) → Business Rules → Semantic Models → Operations Agent
    DATA SOURCES
    OneLake | Lakehouse | Eventhouse | Mirrored DBs | SharePoint | Blob | SAP | Web
    No additional Fabric licensing SKU for IQ (CUs apply). Azure AI Search required for Foundry IQ. Agent Factory P3 for unified metered pricing.
    Resources & Links
    Documentation
    Fabric IQ Overview
    learn.microsoft.com/en-us/fabric/iq/overview
    Foundry IQ Overview
    learn.microsoft.com/en-us/azure/foundry/agents/concepts/what-is-foundry-iq
    Ontology Overview
    learn.microsoft.com/en-us/fabric/iq/ontology/overview
    Tutorials & Hands-On
    Ontology Tutorial (Lakeshore Retail)
    learn.microsoft.com/en-us/fabric/iq/ontology/tutorial-0-introduction
    Agentic Retrieval Tutorial
    learn.microsoft.com/en-us/azure/ai-services/agents/tutorials/agentic-retrieval
    Connect Agent to KB
    learn.microsoft.com/en-us/azure/foundry/agents/how-to/tools/knowledge-retrieval
    Solution Accelerator
    microsoft.github.io/agentic-applications-for-unified-data-foundation-solution-accelerator
    Blog Posts & Sessions
    Fabric IQ Blog (Ignite 2025)
    blog.fabric.microsoft.com/.../introducing-microsoft-fabric-iq
    Foundry IQ Tech Community
    techcommunity.microsoft.com/.../foundry-iq-unlocking-ubiquitous-knowledge
    Ignite Session BRK196
    ignite.microsoft.com/en-US/sessions/BRK196
    Agent Factory
    techcommunity.microsoft.com/.../introducing-microsoft-agent-factory
    Questions & Discussion
  15. Fabric IQ = business meaning (ontology, graph, semantic models, operations agents)
  16. Foundry IQ = knowledge retrieval (knowledge bases, agentic retrieval, single API)
  17. Together = agents that act with intelligence + enterprise security built in
    Treb Gatte | Marquee Insights | Microsoft Data Platform / AI Platform MVP
    Sound off.
    The mic is all yours.
    Influence the product roadmap.
    Join the Fabric User Panel
    Join the SQL User Panel
    Share your feedback directly with our
    Fabric product group and researchers.
    Influence our SQL roadmap and ensure
    it meets your real-life needs
    https://aka.ms/JoinFabricUserPanel
    https://aka.ms/JoinSQLUserPanel
    How was
    the session?
    Complete Session Surveys in
    for your chance to WIN
    PRIZES!