FabCon 2025: Microsoft expands data agent capabilities, reveals OneLake security

MSDW Insight: Fabric’s data agents could serve as key building block for Microsoft’s agent vision, solidifying the platform’s role even as it already sees strong growth.

As the Microsoft Fabric Community Conference 2025 began, Microsoft announced that the customer base has grown nearly 75 percent in the last year, from 11,000 to over 19,000 today. 

As part of this week’s event, the company announced Fabric data agents, which build on the existing AI skills for the platform, and the upcoming preview of OneLake security.

Fabric data agents: Adding analytical resources

In announcing data agents, Microsoft CVP for Azure Data Arun Ulag wrote in a new blog post: 

Fabric data agents not only retrieve data from OneLake, but they can reason over and understand the data—what it means, how it’s structured, and when it’s relevant.

Microsoft will enable Copilot and AI capabilities for all Fabric SKUs “within the coming weeks,” according to Ulag. 

Speaking to MSDW, Ulag said that Microsoft expects Fabric users to rely on data agents for both data retrieval and deep research. 

So what we expect is that businesspeople will have lots of data agents at their disposal based on the different domains that they need to understand better. 

Data agents will build on the semantic models that users already develop in Power BI, he explained, allowing customers’ LLMs to be grounded in not just structured data but relationships. He compared the process of adding Fabric data agents to onboarding a new human data analyst who must become familiar with a range of data sets focused on a business domain before they can deliver insights. That starts with the existing semantic models:

The semantic model is incredibly helpful for the large language model to be able to work with data because the business-friendly name helps them understand how businesspeople refer to the data.

And the data agent can then grow its understanding of a domain before it starts being used in analysis, he noted:

So you [add a new] new data agent, you point the agent to the data. It should work in that particular domain. Give it some helpful instructions, explain what the data [is about]. Give them some example queries. It's very similar to onboarding a business user. And once you create a data agent, you just share it with business people just like you share the Power BI report or a semantic model. The data agent is just yet another artifact that you.

Data agents will become a component of other agents, with a range of potential ways they will fit into a company’s analytics needs. They can be used in the development of other agents developed in Copilot Studio or Azure AI Foundry, they could be consumed as an API endpoint, and they could be consumed via Microsoft 365 applications or elsewhere. 

Fabric’s growth

Fabric’s growth makes it the fastest growing analytics platform in the company’s history. And the company is tracking adoption and uptake in multiple ways. Ulag explained: