What is an agent harness, anyway? A six-minute explainer on how AI coding tools work
This short article explains the parts underneath AI coding tools like Claude Code and Codex. An AI model on its own only produces text. A harness is the software around it that supplies tools such as reading files, running commands or searching the web, then runs whatever the model asks for. The author uses a kitchen comparison: the model is the chef and the harness is the kitchen. It also covers MCP (Model Context Protocol), a standard way to let an AI reach outside services like your email or site analytics.
- For
- People getting into tech who already use AI chat or coding tools and keep running into words like agent, harness, tools and MCP without a clear definition. No coding experience is needed to follow it, though a few code-style examples appear.
- Cost
- Free to read as a public article on X. The author also mentions a paid Mega.dev course and his own open-source product, T3 Code; neither is required to understand the article.
- Time
- 6 minutes, per the article's stated read time
- Access
- Open to anyone with the link. It is posted as an article on X; the page shows log in and sign up prompts, so some readers may be asked to sign in to view it in full.
Worth six minutes if you use AI coding tools and want the vocabulary to finally click. The catch is that the author runs T3 Code and plugs it plus a paid course near the end, so treat those parts as promotion and skip it if you have never touched an AI coding tool yet.
Your moveRead the article once, then try to explain in one sentence what an agent is (an AI model plus a harness of tools) before you open an AI coding tool again.
Start here