Microsoft recently released Fabric IQ, a new product that functions as a semantic layer between data and its consumers. In this article, I discuss what Fabric IQ is and how it helps your organization to build better quality AI agents with greater ease.
AI agents are applications or workflows that use generative AI models, such as large language models (LLMs) to carry out various open-ended tasks. In contrast to conventional applications and workflows, they do not have a pre-determined sequence of tasks to execute in a certain order. Instead, AI agents have a goal that they try to achieve and a set of capabilities that they can use to achieve their goal. The goal is usually defined in natural language and is inserted to the prompt that is used to guide the agent.
The capabilities of AI agents are often called tools and they provide context to the agent, ensuring that the agent’s decisions are grounded on valid and relevant facts. As an example, an agent could be tasked to provide a summary of tasks that each employee in a team has done during a work week. We could provide the agent with capabilities to fetch employee’s work items from a work management system and read employee’s messages from the company’s chat application. This information is then used by the agent to create a summary of the employee’s work week, based on factual and up-to-date information.
Fabric IQ is designed to facilitate the development of AI agents in the Microsoft ecosystem. From the example above we see that integrations to different systems are essential for the agent to get relevant context for the task and make grounded decisions. The purpose of Fabric IQ is to provide built-in tools in Microsoft Fabric for integrating AI agents to the organization’s data platform. These tools help developers to build high-quality AI agents with greater ease and efficiency.
Fabric IQ consists of a semantic layer and built-in agents called Data Agents and Operations Agents. In this article, I will concentrate on the semantic layer that provides agents the data they need to execute their tasks.
Fabric Ontology and the Semantic Layer
In essence, Fabric IQ’s semantic layer is a data modelling and cataloguing tool that is directly connected to the organization’s data estate and offers a programmable interface for AI agents and other software to use. Using Fabric IQ, applications can both understand the meaning (semantics) of data and retrieve the data itself. It also includes features for supervising and improving data quality in the organization, which is often a pain point for large organizations with a complex data landscape.
The central concept in Fabric IQ is the new Ontology feature. Ontology can be understood as a business vocabulary that defines the business entities such as users, sales items or transactions and maps the actual data to the business entities. This “semantic layer” is separate from the data entities and allows to model the business concepts and processes independently from the data schemas defined for tables and files in Fabric Lakehouse and Data Warehouse. The business entities can be constructed as a combination of properties from different tables or by selecting a subset of properties from a single table, for example. Relationships between entities can also be defined in a similar way as in a relational database.
The Ontology also allows to define business logic for data. The business rules and constraints indicate how the data is used in business context, for example by setting accepted ranges for numerical values or accepted values for categorical properties. Furthermore, Ontology allows to define actions to take whenever a business rule is violated. For example, we might want to send an alert to a monitoring system if data contains invalid values or update the value to a correct one. Data quality is often an issue in large organizations where data structures have changed many times over the data lifecycle. The Ontology in Fabric IQ offers a valuable tool for detecting data quality issues and helps to mitigate these issues at the same time.
Note: Business rules feature was not available in our Fabric tenant at the time of writing, but it has been announced by Microsoft in November 2025.
Have I heard this before?
The Ontology is very similar to the “logical data model” or “conceptual data model” concepts used in traditional data modelling. It is more high level than the “physical data model”, focusing on the business understanding of data rather than its technical implementation. The Ontology, however, is directly connected to your data estate and provides both the data and its documentation to end users like product owners, data scientists, data engineers and AI agents.
The Ontology can also be viewed as a data catalog product, since it promotes discoverability of organizational data and includes documentation about the data entities in the organization. This provokes the question, what is the relationship between Ontology and OneLake Catalog or Microsoft Purview? OneLake Catalog and Purview have previously been the primary data catalog and governance services in the Microsoft ecosystem. For the moment, it seems that OneLake Catalog will remain the primary data catalog service in Microsoft Fabric and Purview is the all-round data governance service, while Ontology is directed more towards operational use and AI agent consumption. Perhaps we’ll see some kind of integration between these services in the future, but this is just guesswork for now.
Making relationships with Fabric Graph
In addition to the Ontology, Fabric IQ also includes the new Fabric Graph feature. Graphs are data structures that emphasize relationships between data entities. Graphs may include many different types of relationships between entities, and all these relationships can be retrieved within a single query. This is very powerful when data is highly interrelated or if the data is not readily in a tabular format, for example as a nested object format like JSON.
What makes Fabric Graph stand out from other graph databases is that it is in fact not a database at all. In contrast to conventional graph databases, the data itself is not stored on the graph. The data is stored in Microsoft OneLake and the graph only contains a reference to the actual data. There is no extra step required to load the data from the source system to the graph. This saves an often time-consuming step from the graph creation process.
Fabric Graph can be queried like any other graph database on the market, using the ISO-standardized GQL (Graph Query Language) syntax or natural language. Once data has been ingested to OneLake, it can be added (or “registered”) to the graph and connected to all the other data on the Fabric data platform. It is now one step easier to create an all-encompassing graph for the whole data estate in your organization.
Conclusion
To summarize, the key benefits of Fabric IQ Ontology and Graph features are:
- Enforce consistent definitions and rules for business entities
- Develop AI agents more efficiently, providing the correct context automatically
- Create entity graphs more efficiently, based on existing data in Microsoft Fabric
Fabric IQ is now available as a Public Preview feature. To enable the features discussed in this article, follow these simple steps:
- Set up a Fabric license & capacity and create a Fabric Workspace
- Enable Fabric IQ Preview features for Fabric tenant (instructions)
- Create a Fabric Lakehouse or Data Warehouse and import data to Fabric
- Create an Ontology item in Fabric Workspace and connect your data to the Ontology
- Create a Graph item in Fabric Workspace and connect your data to the Graph
In this article, I have only discussed the Ontology and Graph features in Fabric IQ. Come back soon for the second part of this post series, where I discuss the Data Agent and Operations Agent features and how Fabric IQ powers the agentic AI development for the Fabric data platform in general.
