Microsoft Unifies Agent Development with New Framework
Microsoft has reached a major milestone in its AI agent strategy. The company released its Agent Framework as a Release Candidate (version 1.0.0-rc1) for both .NET and Python, marking the unified successor to its previously separate Semantic Kernel and AutoGen projects.
The framework combines capabilities from both predecessor projects into a single, stable open-source platform for building and orchestrating single or multi-agent AI systems. According to Microsoft's developer blog, the RC release indicates a stable API surface and feature-complete 1.0 release, paving the way for General Availability soon.
What the Framework Offers
Developers can create agents in just a few lines of code with type-safe function tools for calling domain code. The framework supports multiple AI providers including Azure OpenAI, OpenAI, Anthropic Claude, GitHub Copilot, AWS Bedrock, and Ollama.
For orchestration, the framework provides graph-based workflows supporting sequential, concurrent, handoff, and group chat patterns. It includes streaming, checkpointing, human-in-the-loop capabilities, fan-out/fan-in patterns, and hosting on Azure Functions.
Enterprise Features and Interoperability
The framework includes enterprise-grade features: session-based state management, middleware, telemetry, and structured outputs. It inherits Semantic Kernel's type safety and AutoGen's agent patterns while adding explicit workflow control.
Notably, the framework supports emerging interoperability standards including Agent-to-Agent (A2A), AG-UI, and Model Context Protocol (MCP). This positions Microsoft to compete with other major frameworks in the space, including CrewAI (44.3k GitHub stars) and Google's Agent Development Kit (17.8k stars).
Developers can begin production evaluations with the RC release, with full GA documentation forthcoming. Migration guides from Semantic Kernel and AutoGen are in preparation.