What a Harness Is
"Harness" is originally an English word for the straps that secure a horse or a parachute to the body. In software the meaning shifts a little: it refers to the scaffolding you set up in advance so that a tool like an AI works reliably inside a defined frame. In vibe coding, a harness means writing your project's rules, its background, and your own requirements into a file, so you don't have to explain them to the AI every single time.
Think of the onboarding document you hand a new hire on their first day. If you write down "this is what our company does, this folder holds this, don't touch that," the new hire doesn't have to keep asking. A harness is that onboarding document, but for the AI. Because the AI's memory resets each conversation (session), it re-reads this document at the start of every conversation and begins work already knowing what the project is.
Why a Harness Matters Especially for Non-Developers
AI coding tools assume by default that you're a developer, because most of what they learned from is documents, code, and Q&A exchanged between developers. So if you use them with no setup at all, terms like "dependencies," "environment variables," "build," and "deploy pipeline" get thrown at a non-developer with no explanation.
A common misreading here is "if I can't follow it, it must be because I'm not skilled enough." That's not it. The AI simply answers with the safest default (a developer) because it doesn't know who you are. Tell it who you are in a single line, and the same AI starts speaking at a completely different level. A harness is how you lock in that "who I am" once, instead of saying it every time.
What CLAUDE.md and AGENTS.md Are, and Where to Create Them
CLAUDE.md and AGENTS.md aren't special programs - they're just plain text memo files. The .md extension is a format called Markdown, but you can think of it as an ordinary notepad file. You can open and edit it in any text editor, no special software required.
Where you put it matters. This file must sit at the very top (the root) of your project folder. A project folder is the folder that holds one thing you're building - a website, an app, whatever. Create CLAUDE.md in the spot you see the moment you open that folder.
There are two ways to make it. First, just tell the AI to do it. In Claude Code or Cursor, say "create a CLAUDE.md file and note in it that I'm not a developer," and it will make one for you. Second, to do it yourself, create a new file inside the project folder and save it with the exact name CLAUDE.md (capitalization and all).
The First Line a Non-Developer Should Add
Once the file exists, whatever else you skip, start with this one line.
I am not a developer. Always explain technical terms in plain language, as briefly and simply as possible.How That One Line Changes the Answers
That's it. This one sentence tells the AI, at the top of every conversation, "this person is a non-developer," and the AI picks its answers on that premise. When it has to use a term like "environment variable," it explains it in parentheses (a setting value passed to the program), and it trims long-winded explanations down to the essentials. You no longer have to ask "explain it simply" every single time.
From One Line to a Few - a Practical Harness
Once you're comfortable, you can add a few more lines to fit the AI even more to your way of working. It still doesn't need to be longer than a single paragraph.
I am not a developer. Always explain technical terms in plain language, as briefly and simply as possible.
- Before changing code, first explain in one line what you're changing and why, then proceed.
- When telling me to run a command, tell me exactly where (which window) to paste what.
- Before anything risky or hard to undo (like deleting files), always ask me first.CLAUDE.md or AGENTS.md - Which One?
Each line is just "the thing I kept asking for over and over," written down once. A harness grows exactly like this - every time something in the conversation annoys you, move it into the file one line at a time. You don't have to write it perfectly from the start.
So which should you use, CLAUDE.md or AGENTS.md? They play the same role: both are "a project brief you give the AI." The difference is which tool reads them. CLAUDE.md is, as the name says, the file Claude (Claude Code) reads; AGENTS.md is a vendor-neutral shared standard, so other AI tools like Cursor and Codex read it too.
One practical tip: managing two identical files is a chore, so many people put the content in AGENTS.md alone and make CLAUDE.md point to it. But if this is your first time, a single CLAUDE.md is plenty. If you use Claude Code, use CLAUDE.md; if you plan to mix several tools, use AGENTS.md - that's all you need to remember.
What This One Line Saves You
It looks trivial, but the payoff compounds with every conversation. First, the chore of re-asking "explain it simply" every time disappears. Second, shorter explanations mean less reading time and fewer back-and-forth rounds where you didn't understand and had to ask again. Third, when the level matches, you make fewer mistakes running the wrong thing because you misunderstood.
Being good at vibe coding is less about writing good code and more about setting the stage so the AI can help you well. And the first brick of that stage is the single line "I am not a developer." With no elaborate preparation, you can start right now: create one CLAUDE.md in the project folder you're working on, and write that sentence.