Source: AutoGen to Microsoft Agent Framework Migration Guide (Microsoft Learn)

Side-by-side mapping of AutoGen AgentChat/Core concepts to Microsoft Agent Framework agents and typed data-flow workflows.

Contents

Summary

Maps model clients, agents, tools, MCP, agent-as-tool, middleware, custom agents and multi-agent teams from AutoGen to Agent Framework. The central change is from AutoGen’s dual model (low-level autogen-core plus high-level Team) to a single typed, data-flow Workflow with request/response pauses and checkpointing.

Key claims

  • AutoGen AssistantAgent is single-turn unless max_tool_iterations is raised; Agent Framework Agent loops on tools by default.
  • AutoGen offered embedded and experimental distributed runtimes; Agent Framework focuses on single-process composition, with distribution planned.
  • AutoGen GraphFlow is control-flow (broadcast messages, conditional transitions); Agent Framework Workflow is data-flow (typed messages routed along edges to executors).
  • Workflow nodes can be agents, functions or sub-workflows.
  • AutoGen Teams had no built-in pause for human input or checkpointing; Workflows add request/response and superstep checkpoints.
  • RoundRobinGroupChat maps to sequential orchestration; MagenticOneGroupChat maps to MagenticBuilder.
  • The guide lists Swarm-style handoff and SelectorGroupChat as future patterns, and Anthropic/Ollama clients as planned, which conflicts with other Agent Framework pages listing them as available.

Feeds

Suggest changes →