Updated · Originally published
All About Agents is our series covering the basics of AI agents and Microsoft technology. Learn more about .
What are Fabric Data Agents?
Microsoft Fabric Data Agents are a new feature in Microsoft Fabric, currently in preview, designed to let users converse directly with their data without relying on tools like Excel, SQL, or prebuilt Power BI reports. Instead of navigating dashboards or searching across multiple systems, users can pose questions in natural language and receive answers generated from data stored in OneLake. This capability aims to reduce friction for roles such as executives, sales leaders, and operations managers who need quick, ad hoc insights.
While the experience may resemble Power BI Q&A, Fabric Data Agents work differently. Power BI Q&A relied on predefined models and DAX within Power BI, often tied to premium workspaces. By contrast, Fabric Data Agents query underlying data sources directly within the Fabric environment and generate responses using large language models. They can be embedded in experiences like Microsoft Teams and other Microsoft surfaces, enabling conversational access to data without requiring a specific report to be built ahead of time.
Setting up a data agent involves connecting to a data source within Fabric (for example, a Lakehouse), selecting relevant tables, and fine-tuning behavior with instructions and sample queries. Instructions provide context about the data and business rules, while sample questions and SQL examples help the agent learn how to respond accurately. Users can then ask questions such as “What were total sales in California in 2023?” and review both a readable answer and the underlying query for transparency and troubleshooting.
There are important limitations to consider. In the current preview, agents support a limited number of data sources (around five) and tables (roughly 25 for best accuracy), and they currently operate only in English. As with most generative AI systems, they can summarize what happened but won’t reliably explain why trends occurred. Organizations should also plan for capacity management: agent usage consumes Fabric capacity, and heavy adoption can throttle not only the agent but also other Fabric workloads such as Power BI reports and pipelines.
Tips and tricks for working with data agents
Successful use of Fabric Data Agents is iterative. Results improve as teams refine instructions, curate cleaner tables, adjust prompts, and expand or simplify data sources over time. It’s best to start small, monitor usage and accuracy, and evolve the setup based on feedback. As Microsoft continues to invest in this area, expect broader integration options across the ecosystem. For now, clean data, thoughtful configuration, and capacity planning are key to getting value from this emerging capability.