Configuring a Data Agent can be challenging. To deliver reliable answers, authors must understand the data model, identify the right tables and relationships, write clear instructions, create representative examples, and test the agent until it meets user expectations. This process can require hours of schema review, prompt refinement, and manual validation.
Build agent with AI in Data Agent is an AI-powered assistant that simplifies how creators build, configure, and refine Data Agents. Creators can use it to set up a Data Agent or improve one that does not return the expected answers. The assistant replaces manual testing with guided, conversational configuration.
The challenge: configuring Data Agent can be difficult
Data Agents in Microsoft Fabric enable business users to ask natural-language questions about organizational data. Reliable answers depend on defined agent instructions, data source instructions, descriptions, and example queries. These configurations teach the agent how to reason over the data.
Getting these configurations right has historically meant:
- Manually exploring schemas to understand table relationships and join paths.
- Writing and rewriting instructions through repeated testing cycles.
- Crafting example queries from scratch without visibility into what patterns work.
- Guessing at the right level of specificity for instructions.
- Identifying and resolving conflicts across configurations. For example, when an example query implies different business logic than the agent instructions.
Build agent with AI changes that.
What is build agent with AI?
Open build agent with AI by selecting “Build agent with AI” on the Data Agent ribbon. Explore your data, generate configurations, validate them against real data, and apply updates without leaving the Data Agent experience.
Think of build agent with AI as an expert partner that understands your schema, learns from your data patterns, and helps you define the rules your Data Agent needs to deliver accurate answers.
Key capabilities
1. Schema exploration: Understand your data in minutes, not hours
Instead of manually navigating table definitions, simply ask Build agent with AI to explore your connected data sources. It will summarize key entities, identify likely join paths, surface important columns, and recommend the fields most relevant to the questions you want your Data Agent to answer.
Example prompts:
- "Explore the schema and tell me which tables are relevant for sales pipeline reporting."
- "What columns should I use to join customers to orders?"
- "Show me the key date fields and how they're used across tables."
2. Query history exploration: Learn from what already works
Build Agent with AI can analyze patterns from previously executed queries including common joins, filters, and aggregation logic. It uses these insights to recommend better example queries and more precise data source instructions. Instead of starting from a blank page, you build proven patterns.
Example prompts:
- "Look at recent query patterns for churn and suggest a few representative examples."
- "Find common joins between factSales and dimension tables and summarize the join keys."
- "Are there standard filters we always apply for active customers?"
3. Generate instructions and example queries: From exploration to configuration in one step
This is where Build agents with AI truly shine. Based on your conversation, schema exploration, and query history, it generates the configuration that drive Data Agent quality:
- Agent Instructions: High-level rules governing how the Data Agent reasons and selects data sources
- Data Source Instructions: Detailed guidance on tables, columns, joins, and data modeling within a source
- Data Source Descriptions: Summaries of what each source contains and how it's organized
- Example Queries: Natural-language questions paired with queries that capture your business logic
Example prompts:
- "Generate data source instructions so the agent uses the correct join keys and date grain."
- "Suggest five example queries for the most common questions about revenue and returns."
- "Rewrite my current agent instructions to be clearer and remove ambiguity."
You review every proposed change before it's applied. The assistant suggests that you decide.
4. Execute and validate queries: Confidence before you commit
Before applying any configuration changes, use Build agent with AI to run read-only queries against your actual data. Confirm that join paths work, verify data availability, and ensure example queries return the expected results.
Example prompts:
- "Run a sample query to verify the join between dimCustomer and factSales."
- "Test the query you proposed and adjust it if it returns duplicates."
Note: Only read-only operations are supported, your data is always safe.
A simple, iterative workflow
The most successful Data Agent setups follow a natural loop:
- Explore tables, columns, and relationships in your data source.
- Learn and surface proven patterns from query history..
- Generate instructions and example queries tailored to your business questions
- Validate by executing queries against real data to test and refine.
- Apply updates and switch to Test mode to validate end-to-end.
Repeat as needed. Each cycle tightens the feedback loop between configuration and results.
Supported data sources
Build agent with AI works with SQL and Eventhouse data sources in this preview, including:
- Lakehouse (SQL analytics endpoint)
- Warehouse
- Mirrored Database
- SQL Database
- Eventhouse
You can attach multiple data sources, but all must be supported types.
Getting started
- Create a Data Agent and add a supported data source with the relevant tables selected
- Select "Build agent with AI" in the Data Agent ribbon to open Build agent with AI
- Start a conversation by describing what you're building or ask a question you want your Data Agent to handle
- Iterate to explore, generate, validate, and apply configurations through natural conversation
- Switch to Test mode to validate your Data Agent with real questions
No additional setup or configuration is required.
What's next
Try Build Agent with AI today and experience a faster, more intuitive way to build Data Agents that truly understand your data.
Use the Fabric Community or your Microsoft account team to share feedback about the build agent with AI preview.
To continue, see About Data Agent in Microsoft Fabric, review Data Agent configurations, and open the build agent with AI documentation.