Microsoft folds its two agent frameworks into one, and orchestration grows up
Agent Framework 1.0 merges AutoGen and Semantic Kernel. After LangGraph and LangChain reached 1.0, the orchestration layer is standardizing on stable, named patterns.
- On April 3, 2026, Microsoft shipped Agent Framework 1.0, merging its two agent toolkits, AutoGen and Semantic Kernel, into a single production SDK.
- It pairs Semantic Kernel's enterprise features, state management, type safety, and telemetry, with AutoGen's orchestration patterns, and adds graph-based multi-agent workflows.
- The supported orchestration patterns are now named and stable: sequential, concurrent, handoff, group chat, and Magentic-One.
- It follows LangChain and LangGraph reaching 1.0 in late 2025. The orchestration layer is consolidating around a few stable, production-grade frameworks.
The part of the agent stack that decides how agents work together just took a big step toward standardization. Microsoft shipped Agent Framework 1.0 on April 3, folding its two separate agent projects into one supported SDK. AutoGen, which came out of Microsoft Research and pushed the ideas about how agents coordinate, and Semantic Kernel, the enterprise-facing toolkit, are now one framework with stable APIs and long-term support.
What shipped
The merged framework keeps the strengths of both parents. From Semantic Kernel it takes session-based state, type safety, middleware, and telemetry, the plumbing enterprises need before they will run anything in production. From AutoGen it takes the orchestration patterns that let several agents share a job. On top of both, it adds graph-based workflows for laying out multi-agent systems explicitly. It runs in Python and .NET, ships under an open-source license, and integrates with Microsoft's cloud for deployment, according to Microsoft's documentation.
Why merging two frameworks matters
For most of the last two years, a team building multi-agent systems on Microsoft's stack had to choose between a research-flavored library and an enterprise one, and hope the choice aged well. One supported framework removes that fork. It also signals that the orchestration layer is past its experimental phase. Vendors do not converge two products into one with long-term support commitments while the underlying approach is still up for grabs. The named patterns tell the same story. Sequential, concurrent, handoff, and group chat are no longer bespoke wiring each team invents. They are becoming a shared vocabulary, the same way our glossary entry on agent orchestration lays them out.
Orchestration grows up
Microsoft is not moving alone. LangChain and its graph-based sibling LangGraph reached their 1.0 releases in October 2025, with durable state and human-in-the-loop approval built in for long-running agents. Put together, the picture is a field settling on a short list of stable frameworks and a common set of coordination patterns. That maturity is what a multi-agent system needs before a business will trust it with real work, and it pairs with the interoperability standards we covered in the two-protocol agent stack. The wiring is quietly becoming boring, which in infrastructure is the highest compliment there is.
Sources: Microsoft Agent Framework 1.0 announcement, Microsoft Learn, LangChain, LangGraph 1.0.