Introduction

Modern AI solutions often rely on multiple agents working together to analyze inputs, make decisions, and take action. In Microsoft Foundry, agent workflows provide a way to orchestrate these interactions using a combination of agents, control flow, and runtime safeguards.

Foundry includes a visual workflow builder that lets you design and test these systems without writing extensive code. Using the canvas, you can define how agents are invoked, how data moves between steps, and how decisions are made based on agent outputs. You can also observe execution paths and inspect intermediate results to understand how your workflow behaves at runtime.

Suppose you're a developer responsible for automating customer support workflows at a growing SaaS company. Your team receives a steady stream of support tickets ranging from billing disputes to API errors and simple how-to questions. Manually reviewing each request doesn't scale, but fully automating responses isn't always safe. Workflows allow you to combine multiple AI agents, conditional logic, and human-in-the-loop escalation. By using agent-driven workflows, you can triage multiple support requests efficiently and at scale while maintaining reliability and control.

After completing this module, you'll be able to:

• Explain how workflow nodes, variables, and agent outputs work together to control execution paths.
• Use structured agent outputs and conditional logic to route requests to the appropriate workflow steps.
• Implement loops (For-Each) to process multiple inputs efficiently within a single workflow.
• Apply human-in-the-loop and escalation patterns to manage uncertainty and low-confidence agent responses.
• Utilize Power Fx expressions to manipulate data, evaluate conditions, and control flow within workflows.