Your Agent Is Not the Model
Summary
The article clarifies the distinction between an agent and a model, outlining a multi-layer architecture: model, inference service, and harness. It provides practical mapping of common tools (Claude, ChatGPT, LangChain) to these layers and explains how the harness shapes inputs and tool usage, influencing overall behavior. A metaphor of building a house is used to illustrate roles, emphasizing precise terminology to diagnose and improve AI-powered systems.