Agent Harness
ELI5 - The Vibe Check
The agent harness is the exoskeleton around a raw model. On its own, an LLM is just a very smart text predictor sitting in a box. Wrap it in a harness, give it hands (tools), eyes (file and terminal access), and rules of engagement, and suddenly it can read your codebase, run your tests, and open a pull request. Same brain, completely different capability.
Real Talk
An agent harness is the surrounding software, meaning tool definitions, system prompt, loop control, permission checks, and sandboxing, that wraps a base LLM and turns it into an autonomous coding or task agent. Claude Code, Cursor, Aider, and Codex CLI are all harnesses built on top of foundation models rather than models themselves, and the quality of the harness often matters as much as the model inside it.
When You'll Hear This
"Claude Code is just a really good agent harness around the model." / "Half our bugs were in the harness, not the model."
Related Terms
Agent Scaffolding
Agent scaffolding is the training wheels and guardrails wrapped around a model so it acts like a competent worker rather than a genius with no discipline.
Claude Code
Claude Code is Anthropic's agentic coding tool that lives in your terminal.
Sandbox
A sandbox is a safe, isolated play area where code can run without affecting anything real.
Tool Use
Tool use is when an AI can call external functions, APIs, or programs to do things it can't do alone.