But it seems like that there is a higher chance to get questions about the Microsoft Agent Framework than about SK nowadays.
After digging deeper and migrating a small multi-agent Hangman game from classic Semantic Kernel (AgentGroupChat style) to Microsoft Agent Framework (preview), the direction became very clear:
Microsoft is consolidating its agent story under one unified framework:
- Semantic Kernel: still alive and useful, but increasingly acts as a developer-friendly layer
- Microsoft Agent Framework: the new foundation (C# + Python, better orchestration, threading, state management)
- Azure AI Foundry: the production/deployment target
And yes, AutoGen is effectively being phased out.
New development, samples, docs, and investment are now focused on Agent Framework. AutoGen is in maintenance mode (bug fixes/security only).
My honest advice for anyone building agentic applications on the Microsoft stack:
- If you're still on classic AutoGen: plan the migration sooner rather than later
- If you're on older Semantic Kernel patterns: move to the newer ChatClientAgent style first
- If you're starting fresh or want maximum future-proofing: go directly with Microsoft Agent Framework
The migration for my Hangman game (two competing guessers + shared game state) took me ~3 focused hours, mostly figuring out thread handling, Ollama quirks, and tool description wording. Once I clarified the function description, both agents started calling GuessLetter() properly and the game came alive. The new abstractions feel cleaner and more maintainable. The agents also look more stable now.
The Microsoft agent landscape is moving fast, but the preferred path is now quite unambiguous.
The source can be found here: AgentFrameworkPractice
What about you? Still using AutoGen? Experimenting with Agent Framework? Already targeting Foundry? Would love to hear your current stack and migration plans 👇