Semantic intelligence is a critical element of actually understanding what data means and how it can be used.
Microsoft is now deeply integrating semantics and ontologies into its Fabric data platform with its new Fabric IQ technology that it debuted at the Microsoft Ignite conference Tuesday.
Fabric IQ is a semantic intelligence layer designed to address a fundamental problem with enterprise AI agents: Effectiveness depends not just on dataset size but on how well data reflects actual business operations. The new technology creates a shared semantic structure that maps datasets to real-world entities, their relationships, hierarchies, and operational context. The semantic layer represents the latest step in Microsoft's data platform strategy, which recently integrated LinkedIn's graph database technology to provide context.
Microsoft is also expanding its data portfolio with a series of new services: Azure HorizonDB, a PostgreSQL-compatible service in early preview, as well as SQL Server 2025 and Azure DocumentDB, which are now generally available.
"When I think about what fabric does for customers, it gives customers a unified data platform so that they don't have to stitch together many, many, many different tools to get to business value," said Arun Ulag, corporate vice president of Azure Data at Microsoft.
Why semantic understanding matters for AI agents
Traditional AI agents struggle with a fundamental limitation: they can see patterns in data but don't understand what that data represents in business terms. An agent might analyze sales transactions without understanding customer hierarchies, seasonal patterns or product relationships. It can query inventory levels without knowing how production lines connect to distribution networks or how supplier relationships affect availability.
This gap between raw data and business meaning is what causes unreliable predictions and poor automated decisions. Ulag explained that Fabric IQ addresses this by providing a semantic layer that captures how organizations actually operate.
This architectural approach differs significantly from retrieval-augmented generation (RAG) and vector database strategies that competitors have emphasized.
While RAG pulls relevant documents to provide context, Fabric IQ creates a persistent semantic graph representing organizational structure, workflows and business logic. Agents don't just retrieve information. They understand relationships like which suppliers provide which products, how production lines connect to inventory systems or how customer hierarchies map to sales territories.
From analytics semantic models to operational ontologies
Microsoft has invested in semantic models for over a decade through Power BI. These models encapsulate business logic and define entities and relationships; they specify metrics and hierarchies; and they connect to diverse data sources across Azure, AWS, Google Cloud, on-premises systems, and SaaS platforms like Dynamics 365.
"We have 20 million semantic models that run in fabric today. Why? Because we built the semantic modeling layer into Power BI. So behind every Power BI report is a semantic model," Ulag said. "These semantic models already encapsulate a lot of the business logic that mirrors what a customer cares about. What is the data that they care about? What are the metrics that they care about? How does the data relate to each other?"
The limitation of these semantic models has been their scope. They worked well for business intelligence, analytics, and visualization, but they only operated within individual reports or departmental boundaries. Fabric IQ removes these constraints.
"However, we've had a gap. These semantic models were only used for BI use cases," Ulag said. "There's a much bigger opportunity out there, which is the opportunity to be able to take these semantic models and upgrade them into a full ontology."
Upgrading the semantic models to ontologies fundamentally changes what organizations can do with business context and meaning. "What does it do if you upgrade them into an ontology? What happens is that now you can connect data across your enterprise," Ulag said.
He explained that the ontology also integrates with real-time data streams. Beyond connecting data, ontologies allow organizations to define operational rules. This combination creates the foundation for operational agents that understand business context at a level that traditional AI systems cannot achieve. Cross-enterprise data connections work together with real-time integration and rule definitions.
Operational agents that understand and act on business operations
Fabric IQ enables a new class of agents Microsoft calls "operational agents." These agents can autonomously monitor data and take action based on the ontology's understanding of business operations.
"We're also introducing something called operations agents in fabric that can watch your data for you, that can watch the rules that you're asking it to monitor. And it can autonomously take action under human supervision," Ulag said.
Ulag provided a supply chain example that illustrates the difference from traditional approaches. An organization can model its supply chain and delivery operations in the ontology. When real-time data shows congestion in part of a city, the operational agent can automatically reroute trucks around the problem.
The ontologies created in Fabric IQ integrate directly with Microsoft's agent development platforms. This provides business context that makes agents more reliable and accurate.
"It really takes the work that we've done in semantic models in fabric with unified data to a completely different level, allowing customers to be able to model their operations and take business actions," Ulag said.
What this means for enterprise AI strategies
There seems to be a need for context engineering to better enable agentic AI.
Semantics and their associated ontologies do just that and more. Context is about understanding why a request is being made, and semantics understand the deeper meaning. For enterprises struggling with AI agent reliability despite large datasets, Fabric IQ represents a fundamentally different approach. It moves beyond scaling compute or fine-tuning models. The critical question is whether business context captured in ontologies would improve agent effectiveness more than traditional optimization paths.
The strategic bet Microsoft is making is clear: Semantic understanding of business operations determines AI agent effectiveness. Access to large datasets alone is not enough. Upgrading existing semantic models into operational ontologies could provide a faster path to reliable agents.
Credit: Source link
