Vibecoding Guide

Vibecoding means building software by telling an AI tool what you want and letting it write the code. You steer, test, and decide. The AI types.

With vibecoding, you can build:

• Websites and landing pages
• Internal tools and dashboards
• Small web apps and mobile apps
• Automations and scripts
• Prototypes that test ideas

You do not have to write every line. You do have to understand what you ship.

Lovable

Builds full web apps from prompts, right in the browser. The easiest place to start.

Claude Code

Anthropic’s coding agent for the terminal, the desktop app, and IDEs. Comes with paid Claude plans.

OpenAI Codex

OpenAI’s coding agent in the ChatGPT desktop app, the CLI, and IDEs. Comes with ChatGPT Plus and up.

Cursor

An AI code editor for hands-on builders.

Grok Build

SpaceXAI’s open-source coding agent for the terminal.

Many beginners start in a browser builder, then switch to a coding agent for more control.

• Who the app serves
• The problem it solves
• The single most important action
• Each screen and what it shows
• The data it stores
• Sign-in requirements
• What stays out of version one

Interview me about my app idea, one question at a time. Then write a one-page spec with a small version one.

Create a CLAUDE.md or AGENTS.md file covering:

• What the app does
• The tech stack
• How to run it and test it
• Code style rules
• What needs your approval

• Request a plan first
• Approve or edit that plan
• Build a single feature
• Run it and click through it yourself
• Commit the change to Git
• Move on to the next feature

Small, tested steps beat one giant prompt every time.

• Describe the user’s goal, not only the feature
• Set acceptance criteria: “Done when…”
• Share screenshots of the result you want
• Paste the exact error messages
• Have the AI explain anything unclear

• Rely on a proven sign-in service
• Pick a hosted database with access rules
• Ensure users see only their own data
• Store secrets in environment variables
• Store only the data you need

• Request automated tests for core features
• Rerun tests after every change
• Try it on a phone
• Enter wrong inputs on purpose
• Watch a friend use it without help

Review this app for security problems: exposed secrets, missing access checks, unsafe inputs, and data that should not be public. Rank the issues by severity, then fix them one at a time with my approval.

For payments, health, or sensitive personal data, also get a review from a human expert.

• Ship to a preview deployment first
• Set environment variables on your host
• Connect your domain
• Test the live version
• Turn on error monitoring

• Use smaller models for simple edits
• Open a fresh session for each new task
• Keep rules files short
• Use free hosting and database tiers while you are small
• Cap spending on every service

• Ask the AI to explain every change
• Read the code behind your core features
• Learn the fundamentals of your stack
• Learn Git
• Fix one bug yourself every week

The best vibecoders understand a little more each week.

• Building everything at once
• Skipping tests and backups
• Pasting API keys into a prompt
• Shipping code you have never run
• Turning off every approval for agents on your main computer

Act as my senior developer and teacher. App: [idea] for [users]. Read the project rules first. Interview me, write a one-page spec, and wait for my approval. Then recommend a simple stack and explain your choice. Build one feature at a time: plan it, build it, test it, explain it in plain language, and commit it. Never put secrets in code, never delete data, and ask me before you install packages or deploy.