Microsoft released Agent Framework on April 11, an open-source framework for building, orchestrating, and deploying AI agents and multi-agent workflows. The framework supports both Python and .NET with consistent APIs, and ships with migration guides from AutoGen and Semantic Kernel, effectively consolidating Microsoft’s previously fragmented agent tooling into a single platform.

What It Does

Agent Framework covers the full spectrum from simple chat agents to complex multi-agent systems with graph-based orchestration. Core capabilities include data-flow workflows with streaming, checkpointing, human-in-the-loop controls, and what Microsoft calls “time-travel” debugging, which lets developers replay and inspect agent decision chains, according to the GitHub repository.

The framework ships in several layers:

  • Core runtime: Agent creation with tool use, middleware, and provider-agnostic LLM support
  • Graph workflows: Data-flow orchestration connecting agents and deterministic functions
  • AF Labs: Experimental packages for benchmarking and reinforcement learning
  • DevUI: An interactive developer interface for testing and debugging agent workflows

Built-in OpenTelemetry integration provides distributed tracing and monitoring across agent sessions. The framework supports Azure OpenAI, open-source models, and multiple LLM providers, with a middleware system for custom request/response processing, per the official documentation.

The Consolidation Play

The most significant signal is the migration guides. Microsoft is providing explicit paths from both Semantic Kernel and AutoGen, its two prior agent-adjacent frameworks that had overlapping capabilities and confused developers about which to use. Agent Framework is positioned as the answer to that question: use this one.

AutoGen, Microsoft’s open-source multi-agent framework, gained significant adoption but lacked the production-grade workflow tooling and .NET parity that enterprise customers needed. Semantic Kernel offered .NET-first development but had a different architecture. Agent Framework merges the strengths of both while adding the graph-based orchestration layer that neither had.

Developer-First, Not Platform-Locked

The framework emphasizes modularity and avoids vendor lock-in. Developers can use Azure OpenAI or swap in any supported LLM provider. Installation is straightforward: pip install agent-framework for Python, dotnet add package Microsoft.Agents.AI for .NET, according to the GitHub README.

Weekly community office hours and a Discord channel signal that Microsoft is treating this as a community-driven project rather than a top-down SDK, per the repository.

Competitive Timing

The release lands in a crowded week for agent infrastructure. Anthropic launched Claude Managed Agents on April 8. LangChain released Deep Agents Deploy on April 10. GitHub shipped its Copilot SDK (also April 11). Each takes a different approach: Anthropic offers managed hosting, LangChain offers open-source managed agents, GitHub offers an embeddable runtime, and Microsoft offers the development framework itself.

For teams building agents rather than renting them, Agent Framework’s combination of multi-language support, graph orchestration, and open-source licensing positions it as the “bring your own infrastructure” option in an increasingly platform-centric market.