As businesses and data professionals seek more efficient ways to extract insights from their data, Microsoft Fabric introduces a new feature: Fabric Data Agents. Now in public preview, these agents simplify data interaction by allowing users to query structured datasets using natural language—eliminating the need for complex query languages like SQL.

In this blog post, we’ll explore the potential of Fabric Data Agents, their use cases, limitations, and how they differ from Microsoft Copilot.

How the Data Agents work

The Fabric Data Agent leverages large language models (LLMs) through Azure OpenAI Assistant APIs to enable natural, conversational interactions with data. Acting as an intelligent agent, it interprets user questions, identifies the most relevant data source—whether it’s a Lakehouse, Warehouse, Power BI dataset, or KQL database—and selects the right tool to generate, validate, and execute the necessary queries. This allows users to ask questions in plain language and receive clear, structured responses—removing the need to write complex queries while maintaining secure and accurate access to data.

Use Cases of Fabric Data Agents

Fabric Data Agents are designed to democratize data access, making it easier for users to retrieve and interpret structured data insights. Some of the key use cases include:

• Conversational Data Retrieval: Users can ask questions in natural language, and Fabric Data Agents return structured responses without needing deep technical expertise.
• Data-Driven Decision Making: Business analysts and non-technical users can extract insights from enterprise data effortlessly using natural language.
• Enhanced Productivity: Eliminates the need for complex queries, saving time.
• Improved Accessibility: Enables seamless interaction across various datasets, ensuring users at all levels can leverage their data effectively.

How Fabric Data Agents Differ from Copilot

Although Microsoft Fabric also integrates Copilot capabilities, there are distinct differences between Fabric Data Agentsand Copilot in Fabric:

Feature	Fabric Data Agents	Copilot in Fabric
Primary Function	Conversational data retrieval	AI-assisted analytics & coding
Interaction Type	Q&A-based queries	Code generation, data visualization
Data Manipulation	Read-only access	Supports analytics & data transformation
Integration Focus	Querying structured datasets	Assists in building reports, models, and notebooks

While Copilot enhances productivity by generating visualizations, code snippets, and reports, Fabric Data Agents focus on making structured data retrieval more intuitive via natural language interaction.

Configuring a Fabric Data Agent

Setting up a Fabric Data Agent follows a process similar to creating a Power BI report—you first design and refine its configuration to ensure it aligns with your needs, then publish and share it so colleagues can interact with the data. The setup involves several key steps:

Selecting Data Sources

Fabric Data Agents can integrate up to five data sources, allowing flexibility in combining lakehouses, warehouses, KQL databases, and Power BI semantic models. A single agent could contain five Power BI semantic models, or a mix such as two Power BI semantic models, one lakehouse, and one KQL database. You have various options to structure your data environment effectively.

Choosing Relevant Tables

Once you’ve selected your data sources, you need to add them one by one and specify which tables will be used by the Fabric Data Agent. This ensures the agent focuses on the most relevant data, optimizing response accuracy and efficiency.

Adding Context for Better Accuracy

Enhancing a Fabric Data Agent’s accuracy requires providing additional context through agent instructions and example queries. These components help the Azure OpenAI Assistant API better understand user questions and determine the most appropriate data source for each query.

Agent Instructions

You can define custom guidelines to help the underlying agent decide which data sources to reference for specific types of queries. Instructions can also include rules and definitions that clarify organizational terminology or preferences for data retrieval.

Example Queries

Providing sample question-query pairs helps the Fabric Data Agent interpret similar queries with greater accuracy. These examples serve as a reference, guiding the agent in structuring responses that align with expectations.

Limitations of Fabric Data Agents

While Fabric Data Agents offer promising capabilities, there are some limitations to keep in mind:

• Public Preview Stage: Since this feature is still in development, Microsoft may refine its functionality based on user feedback.
• Fabric Capacity Requirements: Users must have a paid Fabric capacity resource (F2 or higher) and enable specific tenant settings to use Data Agents.
• Read-Only Access: Currently, Data Agents only support read queries, meaning users cannot modify datasets through conversational queries.
• Dependency on Data Availability: The effectiveness of responses depends on the quality and structure of the datasets linked to Fabric.
• Limited number of sources per agent: While setting up an agent users can select a maximum of 5 sources (Lakehouses, Warehouses, Semantic Models, …) to be considered in the context of the agent.

This list is not exhaustive. The complete list of limitations can be found here.