Microsoft Agent Framework RC: Semantic Kernel + AutoGen Merge Into a Single Production-Ready SDK

Microsoft has shipped the Release Candidate for its Agent Framework — meaning the API is now locked and production evaluation is officially open. This is the moment to pay attention to what the enterprise AI agent stack is becoming.

What’s in the RC

The framework unifies two previously separate Microsoft projects:

• Semantic Kernel — Microsoft’s orchestration layer for LLM-based applications, with strong enterprise integration and plugin architecture
• AutoGen — the multi-agent conversation framework from Microsoft Research, known for agent-to-agent communication patterns

Previously, teams had to choose between them (or use both awkwardly). The RC merges them under:

• agent-framework-core v1.0.0rc1 — core orchestration, planning, and agent lifecycle
• agent-framework-azure-ai v1.0.0rc1 — Azure AI Foundry integration, model selection, telemetry

Both .NET and Python are supported at RC parity. Java and JavaScript are planned but not RC yet.

What “RC” Actually Means for Evaluators

RC means:

1. The API won’t change before 1.0 — code written against RC can be promoted to production without refactoring
2. Production evaluations are appropriate — this is the stage where teams should run it against real workloads
3. Breaking changes from beta are in — if you evaluated earlier betas, read the migration guide

For teams currently using LangGraph, CrewAI, or other open-source orchestration frameworks, this is the moment to run a side-by-side evaluation. Not because Microsoft is better — the tooling is younger — but because the enterprise integration story (Azure, Active Directory, compliance logging) is stronger, and for large organizations that matters.

The Strategic Picture

Microsoft’s trajectory is clear: dominate enterprise AI agent infrastructure the same way Azure dominates enterprise cloud. The playbook:

1. Develop open-source frameworks (Semantic Kernel, AutoGen) with strong community adoption
2. Integrate them with Azure AI services for enterprise value-add
3. Unify into a single SDK with locked APIs for enterprise confidence
4. Connect to Microsoft 365 Copilot, Teams, and the broader Microsoft product suite

Step 3 just happened. Step 4 is the medium-term roadmap.

Open Source Alternatives: Still Valid

LangGraph and CrewAI maintain genuine advantages:

• More mature ecosystems — more tutorials, community patterns, third-party integrations
• Cloud-agnostic — not tied to Azure
• Faster iteration — RC stability is a constraint, not just a feature

The honest framing: if you’re building on Azure and need enterprise compliance features, Microsoft’s RC deserves serious evaluation. If you’re cloud-agnostic or moving fast with smaller teams, the open-source options likely still have the edge.

The market isn’t winner-take-all here. Different organizational contexts favor different choices, and the RC is the right prompt to explicitly revisit which framework fits your context.

---

Source: InfoQ — Microsoft Agent Framework RC: Semantic Kernel and AutoGen Unified (2026-02)