Agent orchestration matures

An artificial intelligence agent is a software worker that can call tools, keep state, and hand tasks to other workers — and Microsoft just pushed that coordination layer into a 1.0 release for both.NET and Python on April 3. (devblogs.microsoft.com) Microsoft said Agent Framework 1.0 is its production-ready release, with stable application programming interfaces and long-term support after a release candidate in February and an initial launch in October 2025. (devblogs.microsoft.com) The framework is built to run single agents or multi-agent workflows, and Microsoft’s documentation says those workflows use a graph engine that routes work across steps with checkpointing, type-safe routing, and human approval points. (learn.microsoft.com) Microsoft also added declarative agents and workflows, which let teams define behavior in YAML or JSON files instead of writing every step in code. Its documentation says that format is meant to make agents easier to modify and share across teams. (learn.microsoft.com) The release folds together ideas from two earlier Microsoft projects: Semantic Kernel, which handled enterprise features such as state, middleware, and telemetry, and AutoGen, which focused on agent abstractions and multi-agent patterns. Microsoft says Agent Framework combines both in one open-source software development kit. (learn.microsoft.com) The interoperability piece is central. Microsoft says Agent Framework supports the Model Context Protocol, which lets agents plug into outside tools, and Agent-to-Agent, or A2A, an open protocol for agents built in different languages or frameworks to discover each other and exchange tasks over Hypertext Transfer Protocol. (learn.microsoft.com) This is landing as developers move from “one chatbot” demos to systems that split work among specialists. Anthropic says Claude Code now writes most of Anthropic’s own code, and that engineers increasingly spend time on “continuous orchestration,” managing multiple agents in parallel. (anthropic.com) The open-source ecosystem is moving in the same direction. Microsoft’s GitHub repository for Agent Framework had about 9,300 stars as of April 14, and it includes concurrent workflow samples plus separate declarative-agent and workflow examples for Python and.NET. (github.com) Outside Microsoft, projects are testing other coordination models. The MateClaw repository describes itself as a Java and Vue “personal AI operating system” with multi-agent orchestration, memory, skills, and Model Context Protocol support, pointing to a more all-in-one approach than Microsoft’s framework layer. (github.com) The immediate question is no longer whether teams will use multiple agents, but which layer will coordinate them: a vendor framework, a coding tool’s own supervisor, or an open operating system. Microsoft’s 1.0 release puts one of those options on stable footing. (devblogs.microsoft.com)