Vibe Coding Checklist

Free guide · Start from zero · Checked against the official docs, October 2026

Start from zero

How do I build my first real app with AI if I've never coded?

Pick a route by how far you want to go. To try an idea today with nothing to install, describe it to a chat app like Claude and share it as an app with a link. To build a real product, set up an AI coding tool (I use Claude Code in VS Code; Codex, Cursor and Gemini CLI work too), write your rules first, then build the smallest version of one thing.

The hard part of a first app isn't the screens. It's the decisions underneath: where data lives, who can see what, what happens when something fails. This path makes those decisions first, so what you build this week is something you can keep building on.

Pick your route

Today

Try the idea in a chat app

Nothing to install. In the Claude app (what I use), describe what you want and Claude builds it as an artifact you can click through; on the Free, Pro and Max plans, Publish gives it a public link anyone can open. ChatGPT and Gemini have their own canvas features for this too. Best for a tool for yourself or a demo to show people.

Outgrow it when you need sign-in, your own database or your own domain.

This week

Prototype in an AI builder

Lovable, Figma Make and Bolt build a working prototype in your browser. Designers already in Figma will feel at home in Figma Make.

When it needs to become real, use the handover plan to move it into a codebase you own. The handover plan

The real thing

Build it with an AI coding tool

The AI works directly in your project's files on your own computer. More setup, but you own everything from day one, and it's where real products end up anyway. The rest of this page walks you through it.

Takes about an hour to set up. Steps below.

Set up an AI coding tool (about an hour)

  1. Pick your AI coding tool

    Claude Code, OpenAI's Codex, Cursor and Gemini CLI all work in your project's files, and the rest of this guide works with any of them. I use Claude Code in VS Code, so that's the setup shown in the next two steps. Using another tool? Follow its own setup guide, then pick up at Git and Node.js.

    How to know it worked: You've picked one, and you know which plan it needs.

  2. Claude Code: get a paid Claude plan

    Claude Code needs a paid plan: the free plan doesn't include it. Pro is the cheapest that does, at $20 a month (or $17 a month billed yearly, as of October 2026). Max starts at $100 a month for much more usage.

    How to know it worked: Your account page on claude.ai shows Pro or Max.

  3. Claude Code: install VS Code

    A free code editor from Microsoft, at code.visualstudio.com. You'll mostly talk to the AI in it, not type code. (Cursor is its own editor, so Cursor users skip this.)

    How to know it worked: VS Code opens.

  4. Claude Code: add the VS Code extension

    In VS Code, open the Extensions view (Ctrl+Shift+X on Windows, Cmd+Shift+X on Mac), search "Claude Code", and click Install. Open it with the Spark icon and sign in with your Claude account. It needs VS Code 1.94 or later.

    How to know it worked: The Claude Code panel opens and shows you're signed in.

  5. Install Git and Node.js

    Git saves every version of your work and talks to GitHub (git-scm.com). Node.js runs most web apps on your computer (nodejs.org, the LTS version).

    How to know it worked: In a terminal (in VS Code: Terminal → New Terminal), git --version and node --version each print a version number.

  6. Make a GitHub account

    Free at github.com. It's where your code will live, and how hosts like Vercel and Netlify put it online.

    How to know it worked: You can sign in to github.com.

  7. Start a project folder with rules

    Make an empty folder, open it in your editor (File → Open Folder), and save the starter AGENTS.md below in it. Claude Code, Codex and Cursor read that file at the start of every session (for Gemini CLI, name it GEMINI.md).

    How to know it worked: Ask your AI tool "What are the rules for this project?" and it repeats them back.

Your starter rules file (AGENTS.md)

Save this as AGENTS.md in your project folder before you ask for anything else (or GEMINI.md for Gemini CLI). It's the difference between an assistant that guesses and one that follows your rules.

# My app

I am not a developer. Explain what you're about to do in plain English before you do it.

## What this is

See docs/plan.md: who it's for, the problem, and the ONE thing it does first.

## Rules

1. Before changing anything, tell me how we'll know it worked, then show me that it did.
2. Build only what's in docs/plan.md. If you think we need more, ask first.
3. Never put a secret key or password in front-end code or a variable the browser can read.
4. If the app saves data, every table gets access rules (row-level security) the day it's created.
5. Permissions are checked on the server. Hiding a button is not a lock.
6. Emails, notifications and payments can't be taken back. Use test mode and my own addresses until I say otherwise.
7. Small changes, saved to Git often, with plain-English messages.
8. When something fails, show me the real error before trying a fix, and tell me how we'll know the fix worked.

## Decisions

(Filled in on Day 2.)

Your first week

One step a day. Each prompt goes into your AI tool. Don't skip a day's check: "it looks fine" isn't one.

  1. Day 1

    Get interviewed, write the plan

    Before any code, let the AI interview you about the idea. The plan is short on purpose: one kind of person, one problem, one thing the app does first.

    Paste into your AI tool

    I want to build an app, and I'm not a developer. Interview me one question at a time about my idea: who it's for, the problem it solves, the ONE thing it must do first, and what it should not do yet. Then write a one-page plan in plain English to docs/plan.md. Don't write any code.

    How to know it worked: docs/plan.md exists, and you'd be happy to show it to someone.

  2. Day 2

    Decide what's underneath, before it's built

    This is the easiest step to skip, and the most expensive to add later. Decide now which of the ten systems your first version needs, and the simplest safe way to do each.

    Paste into your AI tool

    Read docs/plan.md. For each of these, tell me whether my first version needs it and the simplest safe way to do it: data, accounts, permissions, database rules, secret keys, what happens when things fail, what leaves the app (emails, payments, notifications), uploads, hosting and cost, and knowing when it breaks. Add the decisions to AGENTS.md under "Decisions".

    How to know it worked: AGENTS.md has a Decisions section, and anything you don't need yet is written down as "not yet".

  3. Day 3

    Build the smallest version of the one thing

    Only the one thing from the plan. Let the AI pick a simple, popular setup and explain why.

    Paste into your AI tool

    Build the smallest version of the one thing in docs/plan.md that I can click through on my computer. Pick a simple, popular setup and explain the choice in plain English. Before you start, tell me how we'll know it works. Follow the rules and decisions in AGENTS.md.

    How to know it worked: On your own computer, you can do the one thing from start to finish without help.

  4. Day 4

    Put it online

    Save it to GitHub, then put it on the internet with a host like Vercel or Netlify.

    Paste into your AI tool

    Put this project on GitHub as a private repository, then walk me through deploying it to Vercel or Netlify one click at a time. List any settings it needs by name, never their values. Tell me what to test on my phone afterwards.

    How to know it worked: The app works on your phone, at a real web address.

  5. Day 5

    Watch three people use it

    Send the link to three people who match your plan. Watch them try it without explaining anything. Where they hesitate tells you more than any prompt can. Then run the 2-minute check to see what's still missing underneath.

    How to know it worked: You have a list of what confused people, and your score on the readiness check.

Take the 2-minute checkRead the launch checklist

Newsletter · Free

One new prompt every week.

Subscribers get the Fix Kit: for each of the ten systems, a prompt to check it, a prompt to fix it, and a way to confirm it worked yourself.

Common questions

Do I need to know how to code?

No, but you need to know what you want and check that you got it. The AI writes and runs the code; your job is the plan, the decisions, and testing it like a real user would. The rules in AGENTS.md make it explain itself as it goes.

Do I have to use Claude?

No. Claude Code is what I use, but OpenAI's Codex, Cursor and Gemini CLI work the same way, and every prompt and the rules file here work with any of them. Check each tool's own pricing.

Which Claude plan do I need for Claude Code?

A paid plan: Pro is the cheapest that includes it ($20 a month, or $17 a month billed yearly, as of October 2026). The free plan can build and publish artifacts in the Claude app, but doesn't include Claude Code.

What's the difference between a chat app and an AI coding tool?

In a chat app like Claude, the AI builds a self-contained app you can share with a link. An AI coding tool works inside a real project folder on your computer: many files, a database, your own hosting and domain. Start in a chat app to try an idea; move to a coding tool to build the product.

Can I really have something by the end of the week?

Something small, yes: one thing, working, online. A finished product takes longer. The week plan is designed so that what you have on Friday is built on solid ground, not a demo you'll have to rebuild.

Is this official?

No. It's an independent guide by Andrew Adams, checked against Anthropic's, OpenAI's, Cursor's and Gemini CLI's documentation in October 2026. It teaches you to do it yourself; nobody here builds or reviews apps for you.

Sources

An independent guide, not affiliated with Anthropic, OpenAI, Cursor or Google. Prices and menus change; check the sources if something looks different. Written by Andrew Adams. Already have a prototype? Move it into a codebase you own.