Building Multi-Agent Workflows with Microsoft Foundry & Microsoft Agent Framewor

Many agentic applications need more than autonomous reasoning—they also need predictable execution, durable state, and clear points of human control.

In this session, we’ll take a deeper look at Workflows in the Microsoft Agent Framework and how they provide explicit, graph-based orchestration for coordinating AI agents, functions, and business logic in multi-step applications, using Microsoft Foundry for both the AI model and agent hosting.

We’ll explore executors, typed message routing, conditional branches, fan-out and fan-in patterns, streaming events, checkpointing for pause-and-resume scenarios, and human-in-the-loop approval and review, while discussing when explicit workflows are a better fit than letting an agent dynamically decide the next step.

The goal is to give developers a practical mental model for designing agentic systems that balance agent autonomy with workflow control, resiliency, and observability.