Harness
A harness is the software layer wrapping an AI model. It manages context, tools, and steps, turning a raw model into an assistant that can act.
A model on its own only predicts text. The harness decides what it sees, which tools it can call, and in what order it acts. In practice, it assembles the instructions, hooks, skills, sub-agents, and MCP servers around the model. With the same model, two different harnesses produce very different results on the same task.
Coding assistants like Claude Code, or agents that read a site to answer, all rely on a harness. This layer often matters more than the model itself: it decides whether the AI lands on the right information or gets lost in noise.