Microsoft Agent Framework

About

Microsoft merged its two agent lineages — AutoGen's research abstractions and Semantic Kernel's enterprise machinery — into a single framework maintained by the same teams, so organizations no longer choose between them. Core mechanics center on graph-based workflows that connect agents and functions with typed state, checkpoints, and human-in-the-loop nodes, while middleware intercepts agent actions and MCP supplies tools. A batteries-included Harness Agent adds planning, todo tracking, context compaction, and tool approval for long multi-step tasks, and declarative YAML lets agents be defined without code. Distribution follows conventional SDK channels — Microsoft.Agents.AI for .NET, agent-framework on PyPI, and a public-preview Go module — with OpenTelemetry observability and a DevUI for stepping through workflow graphs. The audience is enterprise .NET and Python teams standardizing on Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic, or Ollama; teams migrating from AutoGen or Semantic Kernel use it as the consolidation path.