Microsoft Agent Framework Reaches GA, Consolidating AutoGen and Semantic Kernel

Microsoft Agent Framework reached v1.0 general availability in April after a release candidate in February, completing the merge of AutoGen and Semantic Kernel into a single SDK aimed at .NET and Azure-anchored teams. It is the most consequential platform consolidation in a year that has already seen agent SDKs ship from nearly every major model provider.

What v1.0 actually delivers

• Graph-based workflows with checkpointing and rollback, mirroring the pattern that has made LangGraph the default for enterprise teams that need audit trails.
• A2A and MCP protocol support, lining the SDK up with the emerging interoperability layer for inter-agent communication and tool exposure.
• Streaming and human-in-the-loop primitives as first-class citizens rather than bolt-ons, reducing the integration glue teams have historically had to write themselves.
• Migration paths from existing AutoGen and Semantic Kernel codebases, which Microsoft has been clear is a one-way door — both predecessor projects are in maintenance.

The framework landscape in mid-2026

A year ago the agent framework conversation was a three-way race. It is now a six- or seven-way one, and most of the new entrants are coming from model providers rather than independent open-source projects. OpenAI shipped its Agents SDK in March; Google introduced ADK in April; Anthropic published its Agent SDK alongside Claude 4.6; CrewAI crossed 44,600 GitHub stars and released v1.10.1 with native MCP and A2A support; and LangGraph reached v0.4 with improved state persistence and human-in-the-loop checkpoints, overtaking CrewAI in stars during the first quarter.

What this means for teams choosing now

The honest answer is that the right framework depends on the stack you already have and the failure modes you most care about. Teams already on Azure and .NET have a much easier decision than they did six months ago — Microsoft Agent Framework is the obvious default. Python-first teams shipping into production with audit and rollback requirements continue to land on LangGraph. CrewAI remains the fastest path from idea to working multi-agent prototype. The conventional wisdom — pick the framework that matches your team's debugging style, not the one with the best demo — still holds.

"The interesting question is no longer which framework. It's which interoperability protocols you're betting on, because that's where the lock-in actually lives now."