Microsoft Agent Framework 1.0 Trails CrewAI and LangGraph on GitHub

Microsoft has given .NET developers a single, officially blessed way to build AI agents. Early signs suggest that consolidation alone won't close the gap with the open-source frameworks that already own developer mindshare.

What Microsoft shipped

On April 3, Microsoft released version 1.0 of its open-source Agent Framework. The release folds two previously separate and incompatible SDKs, Semantic Kernel and AutoGen, into one toolkit 1. Semantic Kernel had served enterprise teams that wanted stable connectors and telemetry. AutoGen had drawn researchers interested in multi-agent orchestration. Until now, developers had to choose one and accept its limits 1.

Microsoft calls the 1.0 release production-ready, with stable APIs and long-term support for both .NET and Python 5. The company describes it as an evolution of the project it first showed in October 2025, built on Semantic Kernel's foundations and AutoGen's orchestration ideas rather than a clean break from them 5. The stable core comes with several features still in preview, including DevUI, hosted agent integration, and expanded observability tooling 5.

The consolidation is real. AutoGen's repository now points new users to Agent Framework and says AutoGen will receive only bug fixes and security patches going forward 1. One developer who has tracked the project since its release-candidate stage notes that Semantic Kernel and AutoGen are both in maintenance mode. That leaves Agent Framework as effectively the only first-party option for new agent work in .NET 3.

The adoption numbers

The community figures are less flattering. One comparison puts Agent Framework at about 9.4K GitHub stars. CrewAI has roughly 48.8K and LangGraph about 29.1K 4. That analysis lists a smaller community, documentation that is still maturing, limited TypeScript support, and a best-case experience that leans heavily on Azure as the framework's main weaknesses 4.

GitHub stars are an imperfect proxy for production use. Still, a five-to-one gap against CrewAI is hard to dismiss. The same comparison notes that CrewAI is Python-only and lacks native Azure or Foundry integration 4. That points to a fairly clear split. Microsoft is winning by default inside .NET and Azure shops, while Python-first teams keep looking elsewhere.

What developers are saying

Reactions in the .NET community vary. One commenter called the framework "half baked" and said few people are doing more than experimenting 2. Another replied that perpetual preview status is normal across the AI tooling world and described using Agent Framework alongside Foundry as the "glue" connecting different agent surfaces, including Copilot Studio 2.

Some developers are simply tired of churn. One said they had already been forced through a major refactor from OpenAI's Assistants API to the Responses API and saw little value in adding another abstraction layer 2. A former AutoGen user took the opposite view, saying Agent Framework fixed their pain points and works well for processing mixed-format data files 2.

The more detailed assessments are mixed but generally positive. Tool-calling, orchestration graphs, and the core agent abstraction are described as solid and conceptually close to LangGraph 3. The .NET experience reportedly feels native rather than like a Python port in C# syntax 3. The weaknesses are familiar ones. Documentation thins out beyond hello-world examples, community extensions are young, and most compelling examples online are still written in Python 3.

The bigger problem is the stack

Forbes argues that the framework is not Microsoft's main issue. Agent Framework 1.0 fixes fragmentation at the SDK layer. But Azure's broader agent offering still sprawls across too many surfaces for building, deployment, governance, and distribution 1. By that account, Google Cloud and AWS each offer a more coherent default path from framework to managed runtime 1. The comparison guide supports this picture, counting more than six Microsoft agent SDKs and treating the Foundry Agent Service and M365 Agents SDK as separate options 4.

Our read

Agent Framework 1.0 is a necessary cleanup, and for .NET teams it is now the obvious choice. That matters, because Microsoft's developer base is large and the framework appears to be technically sound.

It does not yet make Microsoft a contender for the wider agent-building audience. That audience is Python-heavy, values documentation and examples, and has already settled on CrewAI and LangGraph. The low star count reflects a framework that arrived late to an ecosystem that formed without it.

Microsoft has two clear tasks ahead. It needs to grow the framework's community beyond its own platform, and it needs to simplify everything that sits on top of the framework. Until both happen, Agent Framework will likely remain the default for Microsoft shops rather than a choice developers seek out on its own merits.