Building a Multi-Agent Workflow in Microsoft Fabric | DEM362
Alexander Wachtel demonstrates how Microsoft Fabric supports dynamic, context-aware multi-agent workflows using Fabric-native data foundations and orchestration concepts.
Overview
The session explains how Microsoft Fabric enables multi-agent scenarios by combining:
- Agent-ready data foundations in Fabric (semantic models, lakehouses, and pipelines)
- Specialized agents that collaborate across tasks such as planning, retrieval, summarization, and execution
- Natural language interaction, automation, and orchestration to coordinate end-to-end workflows
Key concepts covered
The “3 IQs” framing
The talk introduces how Microsoft Fabric and the “3 IQs” work together to enable multi-agent workflows:
- Work IQ
- Fabric IQ
- Foundry IQ
Fabric foundations for agents
The video highlights Fabric components used as the backbone for agent workflows:
- Semantic models for shared meaning and consistent interpretation
- Lakehouses as the data layer for retrieval and context
- Pipelines for repeatable, orchestrated processing and execution
Built-in Copilot vs external Foundry connections
The session contrasts:
- Using built-in Copilot capabilities
- Using external connections to Foundry for broader agent connectivity and orchestration
Tools and integrations mentioned
The demo references a set of tools that can be used in workflows, including:
- Worker queue
- Copilot
- Teams
- Calendar
- Word
Ontology and data agents
The presenter introduces an ontology approach and demonstrates data agents, including a comparison between:
- Ontology-based querying
- SQL query performance
Connecting agents and workflow setup
The session closes with setup and connection steps for multi-agent workflows, including:
- Connecting agents with storage
- Email-oriented workflows
- Using provided templates to accelerate implementation