The Microsoft Agent Framework (MAF) introduces a workflow programming model for composing AI agents and tasks into multi-step pipelines. Workflows are built from typed Executors wired into directed graphs via WorkflowBuilder, supporting sequential, parallel fan-out/fan-in, conditional branching, and human-in-the-loop patterns. The same workflow definition runs in-process (no dependencies) or durably via the Durable Task Scheduler, which adds checkpointing, distributed execution, and observability. Hosting on Azure Functions provides serverless scaling, auto-generated HTTP triggers, and MCP tool exposure so other AI agents can discover and invoke workflows. Additional patterns include conditional routing with AddSwitch, shared state between parallel executors, and sub-workflows for hierarchical composition.
Table of contents
The Workflow Programming Model Copy linkMaking Workflows Durable Copy linkFan-Out / Fan-In with AI Agents Copy linkHosting on Azure Functions Copy linkHuman-in-the-Loop Copy linkExposing Workflows as MCP Tools Copy linkMore Workflow Patterns Copy linkWrapping Up Copy linkSort: