Enter your search text in the box above

Select a result to preview

Unity in Action – Building Enterprise AI on a Unified Semantic Foundation

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

Description

Enterprise AI requires unified semantic foundations where governance and intelligence flow seamlessly. Learn to architect data platform that use Ontology as your semantic backbone, Data Agents for insights, and Operational Agents for action. Through real-world examples, discover how organizations use AI agents to reliably detect, contextualize, attribute, and act on governed data.

Key Takeaways

My Notes

Action Items

Slides

SESSION AGENDA
Unity in Action — Building Enterprise AI on a Unified Semantic Foundation
The Hook
The Market Reality
The Unified Foundation
Agents in Action
Real-World Case Study
Blueprint & Decision
Framework
The Close
Q&A
The Fragmentation Tax
Does any of this sound familiar?
Your BI dashboard says revenue is up. Your ML model says churn is accelerating.
Your CFO trusts neither.
You added a data catalog last year. Then a governance layer. Then an AI copilot.
Each one brought its own vocabulary — and its own version of truth.
Your AI agents are producing answers fast. But nobody can explain where those answers came from.
Confident answers (hallucination) at machine speed.
This isn’t a
technology problem.
It’s a semantic
problem. And it’s
compounding.
Compliance reviews take weeks because lineage is stitched together manually across six systems.
Every audit is a fire drill.
A new data source lands. Your team spends three weeks mapping it to existing definitions.
By the time it’s ready, the business has moved on.
The Point-Solution Trap
ADDING MORE TOOLS
BUILDING A FOUNDATION
New Catalog
One Semantic Layer
New vocabulary added
Shared language across all tools and agents
New Governance Layer
Governance by Design
More process overhead
Rules flow through the data itself
New AI Copilot
Agents Inherit Meaning
More semantic drift
AI understands context from day one
New Dashboard
Plug-and-Play Sources
Another version of truth
New data inherits meaning automatically
The problem isn’t that you need more tools. It’s that your tools don’t share a language.
“ Return ”
Supply Chain
Finance
Marketing
Tax
Product sent back
ROI
Returning customer
Tax filing
AI Agent
Returns are trending at +3.2% QoQ.
✓ Confidence: 94%
AI that responds. Without reasoning.
Confident. Precise. Wrong.
What ‘Unified’ Actually Means
One Semantic Layer
Governance Flows Through
AI Operates on Meaning
All tools, agents, and users share the
same definitions, relationships, and
business rules.
Security, lineage, and compliance
are embedded in the data — not
applied after the fact.
Agents don’t just process raw data.
They reason about business context,
relationships, and rules.
This is what Microsoft Fabric was built to enable.
Ontology: Your Semantic Backbone
Ontology defines the shared meaning of your data across your entire estate. It’s not a data catalog — it’s a knowledge graph of your business.
Customer
Entities, relationships, and business rules
encoded as machine-readable semantics
Site
Every agent, dashboard, and data product
inherits this shared understanding
Order
New data sources inherit meaning
automatically instead of manual mapping
Governance becomes inherent,
not additional overhead
Regulation
Product
Why Ontology Changes Everything
One Definition – Shared Vocabulary
Across every report, model, and agent. No more
conflicting numbers in the board room.
AI Agents That Reason
Agents know what a ‘high-priority customer’ means,
what ‘revenue at risk’ looks like, and which
regulations apply.
Automatic Meaning Inheritance
New data sources plug in and inherit meaning
instantly. No weeks of custom mapping.
Governance by Design
Rules baked into the data itself. Agents can’t ignore
them. Compliance is inherent, not overhead.
Fabric IQ: Conversational Discovery
Fabric IQ — Conversational Intelligence
What was our
CAC in Q3 for
Northeast?
Compare that to
Q2 and show me
the trend.
Customer Acquisition Cost (Q3, Northeast):
$127.40
Source: Marketing Analytics | Governance:
Verified
How It Works
Which regions are
above our $150
target?
Q2: $142.10 → Q3: $127.40 (∙10.3%)
Driver: Improved organic conversion in digital
channels
Southeast ($163.20) and Midwest ($158.90)
Recommendation: Review paid media
allocation
User Asks
Natural language question — no
SQL, no dashboard hopping
Ontology Resolves
Maps question to governed
semantic definitions
Context Applied
Lineage, confidence, and source
attached
Governed Answer
Trusted response the whole org
agrees on
What if every question your CEO asked got an answer the whole organization agreed on?
From Ontology to Agents: The Semantic Handoff
Without ontology, agents are powerful but blind. With ontology, they understand context, relationships, and rules.
Ontology
Semantic
Handoff
AI Agent
Governed
Action
• Semantic definitions
• Context enrichment
• Contextual reasoning
• Workflow trigger
• Entity relationships
• Rule injection
• Anomaly detection
• Compliance report
• Governance rules
• Lineage tracking
• Pattern recognition
• System update
• Business logic
• Permission check
• Correlation analysis
• Audit trail
Now that the foundation is clear…
LET’S TALK ABOUT
THE AGENTS
Data Agents
Operational Agents
Watch. Learn. Discover.
Autonomous insight generation
on governed semantic models.
Act. Adapt. Execute.
Automated action on governed
playbooks and workflows.
Data Agents: Autonomous Insights
Data Agents are your tireless analysts — constantly monitoring your governed data estate for patterns, anomalies, and opportunities.
Semantic Querying
Agents ask questions in the language of the
ontology — not raw SQL
Cross-Domain Correlation
Connects signals across datasets that the ontology
links — finding patterns humans would miss
Contextual Anomaly Detection
Understands what ‘normal’ means for each entity. A
20% spike at Site A is abnormal; the same spike at
Site B during maintenance is expected.
Governed Output
Every insight carries lineage, confidence score, and
source. No hallucination. No guesswork.
Operational Agents: From Insight to Action
Insights are useless if nobody acts on them. Operational Agents close the loop.
Trigger Workflow
Alert Stakeholders
Data Agent
Operational Agent
Detects anomaly
Contextualizes signal
Determines severity
Follows governed playbook
Human-in-the-loop where
required
Autonomous where trusted
Update Systems
File Compliance
The Agent Architecture: Full Reference
Business Outcomes — Speed to Insight | Compliance Automation | Decision Intelligence | Cost Reduction
Data Agents
Operational Agents
Semantic Querying | Anomaly Detection | Cross-Domain Correlation
| Governed Output
Workflow Triggers | System Updates | Compliance Filing |
Stakeholder Alerts
Fabric IQ — Conversational Discovery | Natural Language Querying | Governed Answers | Pattern Detection
Ontology Layer — Semantic Definitions | Entity Relationships | Business Rules | Governance Policies
OneLake — Unified Data Substrate (Lakehouses, Warehouses, Shortcuts, Mirrored Databases)
The Semantic Maturity Model
Where are you today?
Threshold of Transformation
Autonomous
Unified
Modeled
Cataloged
Siloed
No shared definitions.
Manual
reconciliation.
Metadata exists but
isn’t machineactionable.
Semantic models
defined.
Ontology emerging.
Full ontology. Agents on
governed semantics.
Self-healing data estate.
Agents compose
unprompted.
REAL-WORLD CASE STUDY
Satellite-Based Environmental Monitoring
Fireside Chat with Microsoft Leaders
Green Sky: From Detection to Action: The 4-Stage Value Progression
Transforming Satellite Data into Operational Intelligence
Detect
Attribute
Contextualize
Act
Carbon Mapper – Satellite Data