Autonomous Fabric: Agentic Data Quality Using Fabric Data Agents

#fabcon2026 #session #learning #microsoft #fabcon2026-session

Description

Data Quality issues drain engineering time and weaken business confidence in data platforms.

What if Fabric Data Agents could detect DQ issues autonomously, analyse their root cause, and guide engineers directly to a fix?

This session demonstrates the full architecture with a live demo - capturing DQ signals, triggering agent-led investigations, and integrating agent outputs into ETL pipelines.

Key Takeaways

The Data Quality Problem: Manual DQ checks are slow, automated tools miss edge cases and are expensive, root-cause analysis is time-consuming
Fabric Data Agents are conversational Q&A systems using Generative AI — users chat in plain English over OneLake data, agents reason across multiple data sources
Agents can be fine-tuned for specific business areas (e.g. a DQ monitoring agent that knows your schema, business rules, and expected data ranges)
Under the hood: user question → Azure OpenAI Assistant APIs → LLM generates SQL/KQL → executes against OneLake → returns human-readable answer
Autonomous DQ workflow: Agent continuously monitors data, detects anomalies, runs root-cause analysis, and alerts or auto-remediates without human intervention
Agents can be consumed inside and outside Fabric — integrate into Teams, Power Apps, or custom apps via API
Gotchas: Agent accuracy depends on good schema documentation and instruction quality; always validate generated queries before production use
Speaker: David Mills, Azure Data Architect at ArcAzure — linkedin.com/in/david-mills-etl/

My Notes

Action Items

Resources & Links

Slides

Autonomous Fabric
Agentic Data Quality Using Fabric Data Agents
Author:
David Mills
Position: Azure Data Architect
Email:
dmills@arcazure.co.uk
LinkedIN: linkedin.com/in/david-mills-etl/
David Mills
Azure Data Architect
•
Involved with data since 2000
•
Started career using Oracle products
•
Jumped ship to Microsoft products in 2012
•
Moved into Consultancy in 2019
•
Heavily involved in Microsoft Azure
•
Architecting Microsoft Fabric solutions
•
Plenty of AI & Fabric Data Agent exposure!
•
Now Contracting in 2026
Session Overview
What You’ll Learn Today
•
The Data Quality Challenge
•
Fabric Data Agents 101
•
Agent Configuration
•
DQ Agent Architecture
•
DQ Agent Demos
•
Gotchas & Key Takeaways
•
Q&A
The Data Quality Challenge
• Manual checks slow engineers down
• Automated DQ tools on the market?
• Miss some DQ issues
• Get it wrong!
• Are Expensive
• Root-cause analysis is timeconsuming
• Hard-to-diagnose anomalies
• Poor DQ = Reduced business trust
• Can Fabric Data Agents Help?
What are Fabric Data Agents?
Conversational Q&A
systems that use
Generative AI
Users have plain
English chats over
OneLake data
Leverages Machine
Learning to
understand data
context and
relationships
Agents can be
consumed inside
and outside of
Fabric
Fine-Tune Agents for
specific business
areas
Reason over
multiple data
sources
Quickly find and
extract business
value
Provide immediate
answers or alerts on
your data
How Do Fabric Data Agents Work?
User Asks Agent a
Question in Plain
English
Generates and
Returns Answer in
Human Readable
Format
Applies Azure Open AI
Assistant APIs
•
•
•
Use LLMs for User
Data Extraction
User Permissions (Entra ID)
Security Protocols
RAI Policies
Agent Validates
Question, Generates
and Executes the
Code
Agent Chooses
Relevant Data
Sources and Tables
Fabric Data Agents
Conversational AI Data Quality Architecture
How many rows
have errors?
Connect
Lakehouse
AI
Insights
3 Interrogate
Warehouse
…
Fabric
Data Agent
There are currently
120 rows in the
exceptions table
Question
Engineer
Answer
PBI Semantic
Model
Conversational Data Quality Agent Build
Create Data
Agent
Select Data
Source and
Tables
Configure Agent
• AI Instructions
• Example Queries
Publish & Share
Agent
Consume &
Chat
Fabric Data Agents
API ETL Integration Architecture
ETL – SILVER LAYER
DQ Process
Lakehouse
4*
SILVER_DIM1
BRONZE LAYER
BRONZE
TABLES
Check DQ Rules
and Results 3
Interrogate
(Azure Open API Call)
Q&A Chat
2 Process
*
SILVER_DIM2
SILVER_FACT1
Connect
Populate
Engineer 1
Send Message
Teams Channel
Create & Store
DQ AI
Report
Engineer 2
Review
Interrogate
OneLake
Data Scientist
Fabric Agent API ETL Integration
Insert bad
rows into Silver
Lakehouse
Run Silver
Layer ETL
Trigger Fabric
Agent API &
Interrogate Data
Agent generates
diagnostics &
Fixes
Output Agent
diagnostics
to OneLake
Alert Data
Engineers
Fabric Agent Pre-Requisites and Limitations
• Fabric F2 Capacity or PBI PPC P1 or Higher
• Enable Agent Tenant Settings in Admin Portal
• Need at least 1 and a maximum of 5 data sources
• Only SQL, DAX and KQL languages are supported
• Agent code execution is strictly READ-ONLY
• Unstructured data NOT supported
• Lakehouse files NOT supported
• CANNOT change underlying LLM
• Conversational agent history may NOT persist
• Agent is Region Specific
Next Steps
•External Platform Integration
•Automated Pull Requests
Key Takeaways
• Agent Configuration is Key
• Embed agents within your ETL
• Build Your Own Fabric Agent
• (it’s not that hard!)
Thanks for attending the
session!
Sound off.
The mic is all yours.
Influence the product roadmap.
Join the Fabric User Panel
Join the SQL User Panel
Share your feedback directly with our
Fabric product group and researchers.
Influence our SQL roadmap and ensure
it meets your real-life needs
https://aka.ms/JoinFabricUserPanel
https://aka.ms/JoinSQLUserPanel