Microsoft Ships Agent Framework 1.0, Merging Semantic Kernel and AutoGen into a Single Production-Ready SDK
Microsoft Agent Framework 1.0 unifies Semantic Kernel's enterprise foundations with AutoGen's multi-agent orchestration into one open-source SDK for .NET and Python, with support for A2A and MCP protocols.
Overview
Microsoft released version 1.0 of its open-source Agent Framework on April 3, bringing together two previously separate AI development toolkits into a single, production-ready SDK for building and orchestrating multi-agent systems. The release, announced by Principal Group Product Manager Shawn Henry, marks the framework’s transition from release candidate to general availability with stable APIs and a long-term support commitment.
The framework merges the enterprise-grade foundations of Semantic Kernel with the multi-agent orchestration patterns pioneered by AutoGen, a Microsoft Research project. Developers who had to choose between the two now have a unified programming model available for both .NET and Python.
What We Know
Microsoft Agent Framework 1.0 ships with built-in support for seven model providers: Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, Google Gemini, and Ollama, according to the release announcement. The cloud-agnostic design means agents are not locked into the Azure ecosystem.
The stable API surface covers five orchestration patterns carried over from AutoGen’s research roots: sequential, concurrent, handoff, group chat, and Magentic-One. All patterns support streaming, checkpointing, human-in-the-loop approvals, and pause/resume for long-running workflows, according to the release announcement.
Interoperability is handled through two emerging protocols. The Model Context Protocol (MCP) enables dynamic tool discovery and invocation at launch, while Agent-to-Agent (A2A) 1.0 support for cross-runtime agent collaboration is listed as arriving soon, per the release announcement. Declarative YAML-based agent definitions allow teams to version-control their agent configurations alongside application code.
A pluggable memory architecture supports conversational history, persistent key-value state, and vector-based retrieval, with adapters for Foundry, Mem0, Redis, and Neo4j, according to the release announcement.
The Microsoft Foundry blog describes the framework as resting on four pillars: open standards and interoperability, a research-to-production pipeline, extensible modular design, and enterprise readiness. The enterprise layer includes OpenTelemetry instrumentation for tracing agent actions, Azure AI Content Safety integration, Entra ID authentication, and CI/CD pipeline support for GitHub Actions and Azure DevOps.
Early enterprise adopters named in the Foundry blog post include KPMG for audit automation, Commerzbank for customer support, and BMW for vehicle telemetry analysis, alongside Fujitsu, Citrix, TCS, NTT DATA, TeamViewer, Weights & Biases, and Elastic.
The GitHub repository shows the project has accumulated approximately 9,400 stars and 1,500 forks since its October 2025 debut. The codebase is split roughly evenly between Python (50.5 percent) and C# (45.3 percent), with a small TypeScript component. The project is MIT-licensed.
What We Don’t Know
While A2A protocol support is listed as coming soon, no firm date has been given for when cross-runtime agent collaboration will be fully functional. The DevUI browser-based debugger, Foundry-hosted agent integration, and integrations with the GitHub Copilot SDK and Claude Code SDK all remain in preview, with no announced timeline for stabilization.
It is also unclear how quickly the existing Semantic Kernel and AutoGen communities will migrate. Both predecessor frameworks had significant adoption: AutoGen alone had accumulated over 40,000 GitHub stars before the consolidation was announced. Microsoft has published migration guides for both, but the practical effort required for large-scale projects remains to be determined.
Analysis
The release arrives during a period of rapid consolidation in the AI agent framework space. LangGraph reached 1.0 in October 2025, CrewAI has crossed 44,600 GitHub stars, and both Google and Anthropic have released their own agent SDKs. Microsoft’s decision to merge two established projects rather than maintain them separately reflects the practical reality that fragmentation within a single vendor’s ecosystem creates confusion and slows adoption.
The breadth of model provider support signals a strategic shift. By treating Azure OpenAI as one option among seven rather than a privileged default, Microsoft is positioning the Agent Framework as infrastructure that works regardless of which models an organization selects. Whether that openness persists as the framework matures and competes for enterprise contracts will be worth watching.
For .NET developers in particular, the framework fills a notable gap. Most competing agent frameworks are Python-first or Python-only; the near-equal investment in C# and Python gives enterprises with existing .NET codebases a path into agentic AI without a language migration.