Microsoft Agent Framework is designed to help teams build reliable AI and agent applications with the same engineering discipline they already use in .NET and cloud development. In this conversation, hosts Scott Sauber and Spencer Schneidenbach talk with Shawn Henry, Program Manager for Microsoft Agent Framework, about where the framework came from, why it exists, and how it helps developers move from experiments to production-ready agent systems.

Shawn shares his path through Microsoft—from Windows Phone and Azure Communication Services to Semantic Kernel, AutoGen, and now Microsoft Agent Framework—and explains how those efforts came together to create a more unified approach for building agents. The discussion covers migration from Semantic Kernel, the role of tools, middleware, compaction, memory, and how the framework helps developers avoid the common traps of overly large prompts, tool overload, and brittle orchestration logic.

If you’re working in .NET, Python, or enterprise AI more broadly, this episode offers a practical look at how Microsoft is thinking about agentic development, scale, and the developer experience.

[00:00:00] - Introductions with Scott Sauber, Spencer Schneidenbach, and Shawn Henry
[00:00:19] - Shawn Henry’s Microsoft journey across Windows Phone, Azure, Semantic Kernel, and AI
[00:02:29] - What Microsoft Agent Framework is and the problems it solves
[00:04:30] - Agents, prompts, models, and tools as the core building blocks
[00:06:18] - Migration path from Semantic Kernel and AutoGen to Agent Framework
[00:09:12] - Why Microsoft shifted toward a more agent-centric abstraction
[00:11:46] - Middleware and structured data as underused but powerful capabilities
[00:16:32] - Avoiding too many tools, context overload, and weak agent design
[00:20:29] - Real-world scenarios where agents can add value in existing workflows
[00:24:41] - Dogfooding: how Microsoft teams use agent framework internally
[00:28:37] - VS Code, AI Toolkit, Copilot, and deployment tooling
[00:31:00] - What’s next: GA, enterprise scale, and Claude-like systems

• https://github.com/microsoft/agent-framework

• https://github.com/microsoft/semantic-kernel

• https://github.com/microsoft/autogen

• https://code.visualstudio.com/docs/copilot/ai-toolkit

• https://learn.microsoft.com/en-us/azure/ai-foundry/

• https://aka.ms/agent-framework

Shawn's core message is simple: if you’re already building applications, agent framework helps you extend that code into AI-powered systems without starting over. The conversation also makes a strong case for treating agents as production software—something that benefits from middleware, observability, compaction, and thoughtful tool design.

For teams trying to modernize workflows, reduce toil, or bring AI into existing systems, this episode is a practical starting point.