What is vibe coding?

The principle

You describe what you want in natural language, the AI generates the code, you test and refine. You don't need to know the syntax — but you do need to know how to describe clearly and test what comes out. It's a real shift of skill: the effort no longer goes into writing the code, but into the precision of your thinking.

Concretely, a vibe coding session looks like a conversation. You write "I want a page with a button that adds a habit to a list", the AI produces the corresponding file, you open the preview, you click, and you see what works and what doesn't. Then you rephrase: "the button works, but the list disappears when I reload the page — make it so it gets saved". And so on, until the app does exactly what you wanted.

This way of working has exploded with assistants like Claude, ChatGPT and specialized editors like Cursor. The term "vibe coding" was popularized in early 2025 by Andrej Karpathy, an AI researcher, to describe this way of coding by riding the dialogue rather than typing every line yourself. Behind the laid-back name hides a serious method — the one you're going to learn here.

flowchart LR
  D["You describe in natural language"] --> G["The AI generates the code"]
  G --> T["You test the result"]
  T --> A["You refine the request"]
  A --> D

Why it works now

Five years ago, asking a computer "make me a habit-tracking app" was science fiction. Today, large language models have been trained on billions of lines of public code: they know the recurring patterns of web development better than any human could memorize them. A page with a list, buttons and local storage is a problem they've "seen" tens of thousands of times.

This also explains the limits. The AI excels at what is common and well documented; it is less reliable on what is rare, very recent or very specific to your context. For a first classic web app, you're right in its comfort zone — and that's exactly why this course starts with a simple project.

What it enables (and what it doesn’t)

Start small, you'll scale up in complexity. Tom isn't aiming for an app for 10,000 users: he wants a simple, useful tool, right now. This modesty isn't a flaw, it's a strategy. A small app that's finished and used teaches you a hundred times more than a big project abandoned at 20%.

The boundary between "ideal" and "risky" mostly comes down to the consequences of an error. If Tom's app loses a habit checkmark, it's annoying but harmless. If a banking app loses a transaction, it's a disaster. As a vibe coding beginner, pick projects where the worst possible bug remains trivial: you'll be able to experiment freely, stress-free.

The right mindset

You are the project manager, the AI is your developer. You describe the intention, you check the result, you ask for adjustments. Your value is the clarity of your requests and your testing rigor. A good project manager may not know how to lay bricks, but they know how to tell a straight wall from a leaning one.

This role implies two attitudes. First, unashamed curiosity: you have every right to understand nothing about a line of code and to ask "explain this part to me like I'm ten years old". Second, benevolent skepticism: the AI is sometimes wrong with total confidence. It can claim a feature works when it has never tested it — you are the one who tests, always.

Also expect an emotional rollercoaster. The first generation that works on the first try will give you a feeling of omnipotence; the first incomprehensible bug will make you want to close everything. Both are normal. The vibe coders who succeed aren't the ones who never have bugs, they're the ones who have a method to get through them — you'll learn it in chapter 4.

The classic beginner traps

• Asking for everything at once: a 40-line prompt describing 12 features produces a half-broken app, impossible to debug.
• Never testing: chaining five requests without opening the preview, then discovering nothing works anymore without knowing since when.
• Accepting without reading: approving every AI proposal without even skimming what it says — you lose all ability to understand your project.
• Changing your mind every session: starting a recipe app on Monday, a game on Tuesday, a blog on Wednesday. No project ever gets past the draft stage.
• Believing it's magic: the AI is a powerful but imperfect tool; your attention and your method make the difference between a toy and a real app.

Mini-glossary to get started

• Prompt: the message you write to the AI to ask it for something.
• Generate: the AI's action of producing code from your request.
• Web app: an application that runs in a browser, with no installation, reachable through an address (URL).
• MVP (Minimum Viable Product): the simplest version of your app that is already useful.
• Iterate: improving in small successive touches, rather than doing everything at once.
• Bug: app behavior that differs from what you expected.

You'll come across these words throughout the course and in your exchanges with the AI. No need to learn them by heart: come back to this list when a term blocks you, and never hesitate to ask the AI itself "what's an MVP?". Vocabulary comes with practice.

The running project of this course

To make all this concrete, you'll follow Tom from start to finish. Chapter 2: he picks his tool. Chapter 3: he writes the description of his habit-tracking app. Chapter 4: he builds it feature by feature and fixes the bugs. Chapter 5: he puts it online and shares it with his students. At every step, you'll do the same with your own app idea — it's the best way to learn.