From Single Agent to Multi-Agent Workflows in Microsoft Azure AI Foundry

Most AI demos stop at a single agent answering a question. Real business processes don't work that way. They span multiple steps, multiple systems, and multiple decisions, and a single agent will break long before it gets to the finish line.

This session picks up where every other Azure AI Foundry talk leaves off.

We'll rebuild a failing single-agent scenario live as a multi-agent workflow using Foundry's visual orchestrator and YAML pipeline. You'll see how an orchestrator coordinates domain agents, how state persists across steps, and how MCP tools connect your workflow to real enterprise systems, all the way to a deployed, observable production system.

You'll walk away with the following:
- A mental model for when to use single agents vs multi-agent workflows
- A reusable YAML workflow template
- Hard-won lessons on error recovery and cost control that demos never show you