Everyone is building agents. But most developers new to the space make the same expensive mistake: choosing a framework when they needed a harness — or deploying to the wrong Azure target because nobody explained
the difference.

The model is the engine. The harness is the vehicle. Most developers new to agentic AI skip straight to picking a vehicle without understanding what it actually is — or how it differs from the framework that manages a whole fleet of them. Mixing these up leads to wrong technology choices,
hard-to-maintain code, and painful rewrites.
This session untangles both layers from first principles. We start with a beginner-friendly definition of what an AI agent actually is — memory, tools, and an action loop — then map the full landscape of frameworks and harnesses. From there, we compare three Azure-relevant harnesses in depth: the Claude Agent SDK, the GitHub Copilot Agent SDK, and the Microsoft Agent Framework.