Designing AI Agent Workflows on Modern Data Platforms

Hi everyone,

I have been exploring how AI agents can work with modern data engineering platforms to automate business processes.

A common architecture I see is that data pipelines collect and prepare information, AI agents analyze the context and determine the next action, workflow services execute tasks through APIs or connected systems, and the results are stored for reporting or further processing.

I am interested in how teams are approaching this with Microsoft Fabric.

Are you using AI agents directly with Fabric workflows, or keeping the AI layer separate?

What patterns work well for connecting AI agents with Lakehouse, Data Pipelines, or notebooks?

How do you manage security and permissions when AI applications need access to enterprise data?

Are there recommended approaches for combining Fabric workloads with external AI services?

I would be interested to hear what architecture patterns others are using and what challenges you have encountered.