Microsoft merged AutoGen and Semantic Kernel into Agent Framework. The production-grade successor targets .NET and Python developers with graph-based workflows and built-in observability.
Key facts
- Agent Framework merges AutoGen and Semantic Kernel into one.
- Supports graph-based workflows: sequential, concurrent, handoff, group collaboration.
- Built-in checkpointing, streaming, human-in-the-loop, time-travel debugging.
- Deploy to Foundry-hosted infra with two extra lines of code.
- Includes OpenTelemetry tracing and DevUI for testing.
Microsoft has unified its two major AI agent frameworks—AutoGen and Semantic Kernel—into a single offering called Agent Framework, according to a post by @_vmlops on X. The new framework is described as the production-grade successor to both, built by the same teams for .NET and Python.
Agent Framework supports graph-based workflows including sequential, concurrent, handoff, and group collaboration patterns. It includes built-in checkpointing, streaming, human-in-the-loop, and time-travel debugging—features that previously required custom implementation or third-party tools. According to @_vmlops
Key Capabilities and Deployment
The framework introduces declarative agents defined in YAML, allowing developers to specify agent behavior without writing boilerplate code. A DevUI enables testing workflows before shipping to production. For observability, Agent Framework includes built-in OpenTelemetry tracing and debugging.
Deployment is streamlined: developers can deploy to Microsoft Foundry-hosted infrastructure with two extra lines of code, reducing the friction of moving from development to production. This targets the enterprise pain point of complex agent orchestration and infrastructure management.
Strategic Rationale
The merger resolves a long-standing confusion in Microsoft's agent ecosystem. AutoGen, originally from Microsoft Research, focused on multi-agent conversation patterns and was popular in research settings. Semantic Kernel, developed by the Azure AI team, emphasized enterprise integration with existing .NET and Python codebases. Maintaining both created fragmentation and forced developers to choose between research-grade flexibility and production reliability. Agent Framework collapses this choice, offering a single stack that spans prototyping to deployment.
The move mirrors broader industry consolidation as agent frameworks mature. LangChain and CrewAI have also expanded their feature sets, but Microsoft's advantage lies in tight integration with Azure infrastructure and Foundry. By merging rather than deprecating, Microsoft signals that both research and enterprise use cases are first-class citizens in the new framework.
What to watch
Watch for the first production deployments on Foundry in Q2 2026 and whether Microsoft deprecates standalone AutoGen and Semantic Kernel packages. Adoption metrics from enterprise previews will reveal if the unified framework reduces developer friction or introduces new complexity.
