What Is Claude Code (and What Is a "Harness")?
If you’ve used a chatbot, you’ve talked to a model — a system that reads your text and writes text back. That’s useful, but it has a hard limit: a model on its own can’t do anything. It can describe how to rename ten files, but it can’t open the folder and rename them. It can write code, but it can’t run the code, see the error, and fix it.
Claude Code closes that gap. This page explains what it is and the one idea — the “harness” — that the rest of this site builds on.
A model vs. an agent
Section titled “A model vs. an agent”Start with the two words, because people mix them up constantly.
- A model (like Claude Opus 4.8) is the language brain. You give it text, it gives you text. It has no hands.
- An agent is a model plus the ability to take actions in the real world — read a file, run a command, search the web — look at what happened, and decide the next step. It loops until the job is done.
The difference is doing vs. describing. A chatbot describes the fix. An agent applies it, runs the tests, and tells you whether they passed.
What a “harness” is
Section titled “What a “harness” is”The piece that turns a model into an agent is called a harness. The word is borrowed from horses: a harness is the rig that connects a powerful animal to the cart so its power becomes useful work. Same idea here. The model is the power. The harness is the rig that points that power at a real task.
A harness gives the model four things:
- A loop. The model doesn’t answer once and stop. The harness runs it again and again — act, observe the result, decide what’s next — until the task is finished or it gets stuck.
- Tools. Concrete actions the model can call: read a file, write a file, run a terminal command, search your code. (We cover these one by one in Tier 1.)
- Memory and state. A way to remember things that matter — usually by reading and writing files on disk, not by keeping everything in the conversation.
- A way to talk to you. It shows you what it’s about to do, asks permission for risky actions, and reports results.
Claude Code is a harness. It’s a program you run in your terminal that wraps the Claude model in exactly that loop-plus-tools-plus-memory rig. You give it a goal in plain English; it figures out which tools to use, in what order, and works the problem.
A concrete example
Section titled “A concrete example”Suppose you tell Claude Code: “There’s a typo in the heading on my About page — fix it.”
A plain chatbot would reply: “Open the file, find the heading, change the word, save.” Now it’s your job.
Claude Code, the harness, does this instead:
- Uses its search tool to find files that look like an About page.
- Uses its read tool to open the most likely one and look at the heading.
- Uses its edit tool to change the typo — and shows you the exact before/after first.
- Tells you it’s done, and where.
Same model. The difference is the harness around it — the loop, the tools, and the file access.
Why this matters for the rest of the site
Section titled “Why this matters for the rest of the site”Almost everything advanced you’ll read here — agents that run on a schedule, agents that talk to each other, a self-healing system that reads its own error logs and proposes fixes — is just this one idea, scaled up. It’s still a model, in a loop, using tools, keeping its memory in files.
If you hold onto one sentence from this page, make it this: a harness turns a model that can only talk into an agent that can do.
Next: You understand what Claude Code is. Now install it so you can follow along.