Data Architect BI Developer Business Decision Maker Platform Owner — This section introduces Fabric IQ, Microsoft's semantic intelligence layer that extends Fabric beyond data management into enterprise-grade AI-driven operations.
Microsoft Fabric IQ
From data platform to intelligence platform — the semantic layer that enables AI agents to reason over, decide, and act on your business data.
What is Fabric IQ?
Microsoft Fabric IQ is a semantic intelligence layer that transforms Fabric from a unified data platform into a comprehensive intelligence platform. Rather than just storing and visualizing data, Fabric IQ organizes business concepts, rules, and operational logic so that both people and AI agents can interpret, reason about, and act on data in real time.
Announced at Microsoft Ignite 2025, Fabric IQ addresses a core enterprise challenge: fragmented business logic, inconsistent metrics between departments, and limited real-time decision-making capabilities.
Fabric IQ represents a paradigm shift: your data platform doesn't just store data — it understands it. Trusted Power BI semantic models are extended beyond analytics into operations and automated decision systems, creating a shared vocabulary that both humans and AI agents use.
Core Components
Ontology Project
Visual, no-code tools for building business ontologies — models of entities (customers, orders, products), relationships, workflows, and rules. Creates a shared vocabulary across the organization.
Unified Semantic Layer
All applications — human and AI — leverage consistent definitions and reasoning over data. Extends Power BI semantic models into the operational layer, ending metric fragmentation.
Graph Engine
Enables multi-hop reasoning for deep, cross-domain insights. Connect entities across business domains to uncover hidden relationships and dependencies.
AI Agents
Data Agents (now generally available) answer complex business questions based on shared meaning. Operations Agents act autonomously — interpreting live signals, making decisions, and taking action.
Fabric IQ Architecture
Ontology: Modeling Your Business
The Ontology Project is the foundation of Fabric IQ. It lets business experts define how the organization works — without writing code. An ontology captures:
| Concept | Description | Example |
|---|---|---|
| Entities | Core business objects that your organization tracks | Customer, Order, Product, Employee, Store |
| Properties | Attributes of entities, mapped to data columns | Customer.LifetimeValue, Order.Status, Product.Category |
| Relationships | How entities connect to each other | Customer → places → Order → contains → Product |
| Rules & Logic | Business rules and workflows that govern decisions | "High-value customer" = LTV > $10K and orders > 5/year |
| Metrics | Organization-wide metric definitions (single source of truth) | Revenue = SUM(Order.Total) WHERE Order.Status = 'Completed' |
Start your ontology with existing Power BI semantic models — they already contain trusted metric definitions, relationships, and business logic. Fabric IQ extends these models into the operational layer, so you don't start from scratch.
Example: Ontology Definition
{
"ontology": "RetailOps",
"entities": [
{
"name": "Customer",
"source": "Lakehouse.dbo.dim_customer",
"properties": [
{ "name": "LifetimeValue", "type": "currency", "column": "ltv_amount" },
{ "name": "Segment", "type": "string", "column": "customer_tier" },
{ "name": "ChurnRisk", "type": "decimal", "column": "churn_score" }
],
"rules": [
{ "name": "HighValue", "condition": "LifetimeValue > 10000 AND OrderCount > 5" }
]
},
{
"name": "Order",
"source": "Warehouse.dbo.fact_orders",
"properties": [
{ "name": "Total", "type": "currency", "column": "order_total" },
{ "name": "Status", "type": "string", "column": "order_status" }
]
}
],
"relationships": [
{ "from": "Customer", "to": "Order", "type": "one-to-many", "verb": "places" }
],
"metrics": [
{ "name": "Revenue", "expression": "SUM(Order.Total)", "filter": "Order.Status = 'Completed'" }
]
}
AI Agents in Fabric IQ
Fabric IQ introduces two types of AI agents that operate on the semantic intelligence layer:
Data Agents
Answer complex business questions using natural language, grounded in the ontology's shared definitions. Unlike generic chatbots, Data Agents understand your specific business context.
- "What's our customer churn rate by region this quarter?"
- "Which product categories drive the most repeat purchases?"
- "Compare revenue vs. forecast across all business units"
Operations Agents
Act autonomously on live business data — they monitor signals, interpret events, make decisions, and take operational actions, continuously learning and improving.
- Detect inventory anomalies and trigger reorder workflows
- Identify at-risk customers and create retention campaigns
- Monitor SLA breaches and escalate to appropriate teams
Data Agents vs. Operations Agents
| Aspect | Data Agents | Operations Agents |
|---|---|---|
| Purpose | Answer questions & generate insights | Monitor, decide, and take autonomous action |
| Trigger | User asks a question (on-demand) | Live signals and business events (continuous) |
| Output | Answers, reports, visualizations | Workflow actions, alerts, record updates |
| Interaction | Conversational Q&A | Autonomous with human escalation |
| Governance | RBAC + semantic model access | Policy controls + approval thresholds + audit trail |
| Example | "What's our churn rate by region?" | Auto-reorder when inventory drops below threshold |
| Status | Generally Available (FabCon 2026) | Preview |
Data Agent → Your team needs answers from data but shouldn't need to write SQL or DAX. Start here — it's GA and lower risk.
Operations Agent → You have a repetitive operational workflow that follows clear business rules (e.g., inventory alerts, SLA monitoring). Requires governance setup and human-in-the-loop configuration.
Both together → A Data Agent surfaces an insight ("churn is spiking in EMEA"), and an Operations Agent acts on it (triggers a retention campaign).
Example: Querying a Data Agent
from azure.identity import DefaultAzureCredential
import requests
credential = DefaultAzureCredential()
token = credential.get_token("https://api.fabric.microsoft.com/.default")
# Query a Data Agent grounded in your ontology
response = requests.post(
"https://api.fabric.microsoft.com/v1/workspaces/{workspace_id}/agents/{agent_id}/query",
headers={
"Authorization": f"Bearer {token.token}",
"Content-Type": "application/json"
},
json={
"question": "What are the top 5 products by revenue this quarter?",
"context": {
"ontology": "RetailOps",
"entities": ["Product", "Order"],
"timeframe": "current_quarter"
}
}
)
result = response.json()
print(result["answer"]) # Natural language answer
print(result["data"]) # Structured data behind the answer
print(result["lineage"]) # Data sources and semantic models used
Real-Time Operational Intelligence
Unlike traditional BI that focuses on after-the-fact reporting, Fabric IQ enables real-time operational intelligence. Operations Agents can:
- Monitor live signals: Stream data from Eventhouse, databases, and external systems in real time
- Interpret with context: Use the ontology to understand what events mean in business terms (not just raw data)
- Decide and act: Trigger automated workflows, send alerts, update records, or escalate to humans
- Learn and improve: Continuously refine decisions based on outcomes and feedback
Governance & Human-in-the-Loop
Fabric IQ is designed with enterprise governance at its core. AI agents don't operate unchecked:
- Policy controls: Define what agents can and cannot do — scope their access to specific entities, actions, and data domains
- Human-in-the-loop: Configure approval thresholds — agents can act autonomously for routine decisions, but escalate high-impact actions for human review
- Audit trail: Every agent decision and action is logged and traceable, integrated with Purview compliance
- Sensitivity labels: Agents respect the same MIP sensitivity labels applied to Fabric data — they can't access or expose data beyond their authorized scope
While Data Agents reached general availability at FabCon 2026, the Ontology Project and several Fabric IQ capabilities remain in preview. Features, APIs, and capabilities are evolving. Start experimenting with pilot scenarios and existing semantic models, but plan for changes as the platform matures.
What's New — FabCon 2026
Microsoft unveiled major Fabric IQ advancements at FabCon 2026 (March 2026), moving key capabilities from preview to general availability and introducing new planning and automation features.
Planning in Fabric IQ
Native forecasting and budgeting workflows run directly on top of existing Fabric SQL and semantic models. Writeback operations store planning data in regular Fabric SQL tables, instantly available to BI and analytical processes — eliminating separate planning databases or complex integrations.
- Budget planning with direct writeback to SQL tables
- Forecasting workflows on top of semantic models
- Planning data immediately available for BI reports
Data Agents — General Availability
Fabric Data Agents are now generally available as "virtual analysts." They leverage generative AI and domain-specific knowledge from the ontology to automate data analysis tasks and enable conversational Q&A across all data formats and models in Fabric.
- Generative AI grounded in your business ontology
- Domain-specific knowledge for accurate answers
- Works across Lakehouse, Warehouse, and semantic models
AI & Copilot Integration
Copilots and AI agents now offer deeper integration with automated analysis suggestions, code generation, and deployment workflows. GitHub Copilot operates directly within Fabric environments, respecting RBAC and governance boundaries.
- Automated analysis suggestions in notebooks
- Code generation respecting Fabric RBAC
- GitHub Copilot integration within Fabric workspaces
📚 Learn More
FabCon 2026: New Capabilities to Unify Databases ↗Connected Data Estates
Fabric IQ doesn't operate in isolation — it integrates tightly with the broader Microsoft data ecosystem, enabling the semantic layer to reason across transactional, analytical, and operational data in a unified architecture:
OneLake & Fabric Workloads
Ontology entities map directly to Lakehouse tables, Warehouse views, and real-time streams. The semantic layer provides a unified business vocabulary over all Fabric storage formats.
Database Hub
The new Database Hub provides observability and governance across SQL Server, Azure SQL, Cosmos DB, and PostgreSQL — surfacing operational database data directly into the Fabric IQ semantic layer.
Dynamics 365 & SaaS Apps
Fabric IQ connects operational signals from Dynamics 365 (Sales, Finance, Supply Chain) and other SaaS applications, bridging the gap between business applications and the semantic intelligence layer.
Real-Time Mirroring
Continuous, near-real-time database mirroring into OneLake from operational systems like Oracle, SAP, and SharePoint — making business-critical data instantly available for Fabric IQ reasoning.
Fabric IQ leverages OneLake's zero-copy integration with external platforms (Snowflake, Databricks). Data stays where it is — the semantic layer reasons over it without duplicating or moving it, reducing complexity and cost.
Microsoft IQ Ecosystem
Fabric IQ is part of a broader Microsoft IQ initiative — three intelligence layers designed to power the agentic enterprise:
| Intelligence Layer | Scope | Purpose |
|---|---|---|
| Work IQ | Microsoft 365 | Intelligence across documents, emails, meetings — understands how work gets done and powers Copilot agents |
| Fabric IQ | Microsoft Fabric | Semantic intelligence over business data — ontology, metrics, business rules, and data/operations agents |
| Foundry IQ | Azure AI Foundry | AI model intelligence — model context, grounding, and evaluation for building custom AI solutions |
Getting Started with Fabric IQ
Step 1: Audit Your Semantic Models
Identify your most trusted Power BI semantic models — these are the foundation. Ensure they have clear metric definitions, proper relationships, and endorsement (Certified status).
Step 2: Define Your Ontology
Start small — model 2-3 core entities (e.g., Customer, Order, Product) and their relationships. Use the no-code Ontology Project tools to build your first business model.
Step 3: Create a Data Agent
Build a Data Agent grounded in your ontology. Test it with business questions your teams frequently ask. Validate that answers align with your established metric definitions.
Step 4: Pilot an Operations Agent
Choose a low-risk, high-frequency operational workflow (e.g., inventory alerts, report distribution). Configure guardrails, approval thresholds, and monitoring before expanding scope.