What even is an "agent harness"?

A deep-dive explanation of what an 'agent harness' is in the context of AI coding tools. The harness is the environment and set of tools (read file, list files, edit file, bash) that an AI agent operates within, handling tool calls, chat history, and re-requesting from the model after each tool execution. The video demonstrates building a minimal harness in ~60-75 lines of Python, showing how tool descriptions in the system prompt heavily influence model behavior. It explains why Cursor's harness produces better results than raw model APIs (careful prompt and tool description tuning per model), and clarifies that T3 Code is a UI layer on top of existing harnesses like Claude Code and Codex CLI, not a harness itself.