Every company has them. You know, those go-to people that just know where important data lives, how to interpret the cryptic column names, and which reports to trust (and which ones to take with a grain of salt). This kind of institutional knowledge is invaluable, but it’s also fragile. When those experts take time off, change roles, get hit by a bus, or leave the company entirely, the insight they carry often goes with them.
Enter Microsoft's new Fabric Data Agents - now in public preview.
These intelligent, AI-powered agents are designed to bridge the gap between raw data and real-world context. Rather than starting from scratch every time someone needs a report or insight, Fabric Data Agents can be trained to understand your data landscape. In other words, they understand how legacy systems work, what tables to go to to find answers to specific questions, custom business logic, and more. They learn how your organization thinks about data, so they can help new users (and even seasoned pros) find answers faster, with far less ramp-up.
In this post, we’ll explore how data agents turn tribal knowledge into institutional intelligence and why they’re becoming a game-changer for teams looking to scale their analytics capabilities without scaling their training budget.
Introducing Fabric Data Agents
At its core, a Fabric data agent is an AI-powered assistant that knows your data and knows how to talk to it. Built on top of Azure OpenAI’s Assistant APIs, Fabric data agents bring world-class natural language understanding (NLU) to the data workspace. This means they can take a plain-English question from a user and translate it into structured query languages like SQL, DAX, or KQL, depending on the data source.
But what makes these agents truly powerful isn’t just their ability to understand language—it’s their deep integration with Microsoft Fabric itself. Data agents can analyze and reason over all the data assets you’ve already built in Fabric/OneLake:
Data Warehouses and Lakehouses for structured and unstructured data
Power BI Semantic Models for curated business logic and KPIs
KQL Databases for real-time streaming and log analytics
Once connected to these data sources, data agents can be trained much like a new data analyst. You provide them with context—things like instructions, common business questions, and example queries—and they learn how your data environment is structured, what tables matter, how they relate, and where specific insights can be found - see Figure 1 below for context.
Figure 1: Fabric Data Agents - Basic Concepts
At runtime, users can simply ask questions like “What were our top-selling products last quarter in the Northeast?” and the data agent will do the heavy lifting by:
Translating the request into an executable query.
Running it against the appropriate source.
Applying user security policies and returning a secure, accurate result.
Because these agents operate within Fabric, everything happens within your established security model. The agent knows who the user is, what they’re allowed to access, and ensures that any data retrieval respects those boundaries. In other words: data agents are powerful, but they play by the rules.
Putting Data Agents to Work
Once a data agent is fully trained, it's a bit of a choose-your-own-adventure in terms of how you put it to work. Besides direct use inside of Fabric itself—either directly in Power BI or programmatically—Microsoft has begun integrating with data agents across a wide variety of solutions.
For example, Figure 2 below illustrates how data agents are integrated into Microsoft Copilot. As more and more users get up to speed with Copilot, data agents are a great addition to the list of purpose-built agents being added to Copilot every day.
Figure 2: Working with Data Agents in Microsoft Copilot
As you can see in Figure 3 below, the user experience with data agents in Copilot is the same as any other agent. Even though the data agent is more data-focused, users can interact with it without having any pre-existing knowledge about the underlying data sources.
Figure 3: Chatting with a Data Agent in Microsoft Copilot
We also have the option of incorporating our data agents into larger-scale agent solutions. For example, you can see how we're incorporating a data agent into a custom agent we're defining in Microsoft Copilot Studio. This capability unlocks many powerful low-code mashup solutions like the ones we demonstrated in our blog series on redefining user experiences with Copilot Studio (ref: Part 1, Part 2, and Part 3).
Figure 4: Integrating Data Agents into Microsoft Copilot Studio
Of course, we're not just limited to consumption from low-code tools like Copilot Studio. Microsoft has also incorporated data agents into Azure AI Foundry where we can bring the full power of Azure and OpenAI to bear on building cutting edge agent experiences.
Closing Thoughts
In many ways, Fabric data agents represent the next evolution in how we engage with data. With data agents, data is not just as a resource to be queried, but institutional knowledge to be captured, shared, and scaled. By combining natural language understanding with deep integration into your existing Fabric data estate, these agents offer a way to democratize access to insight without compromising on security or context.
Instead of relying on a few seasoned veterans to explain where the data lives (and what it really means), we can now train data agents to carry that knowledge forward. This investment makes it easier for new employees to ramp up, for business users to get answers, and for data teams to spend less time fielding basic questions and more time building what’s next.
We'll be exploring these concepts in greater detail in the coming weeks as we delve deeper into AI integration across the enterprise.
