Vibe Coding: The Generation Building Apps by Talking to AI (Without Knowing How to Code)
What is vibe coding?
The practice of creating software by talking to an AI in natural language: you describe what you want, test what it generates, request adjustments and repeat, without writing or auditing the code. The term was coined in 2025 by researcher Andrej Karpathy.
People are shipping apps without knowing what a variable is. They describe the idea to an AI in plain language ("I want a site to organize my friends' barbecue schedule"), request tweak after tweak ("now make the button bigger", "now save the history"), and within hours have something working on screen. The practice earned a name: vibe coding, programming by feel, through conversation, without reading (or understanding) the code the AI writes underneath.
It became a joke among programmers and a fever among everyone who ever had an idea stuck in a drawer. Both reactions are right at the same time, and that is what makes the subject interesting.
What it is, exactly
Vibe coding is using an AI as your programmer while you behave like the client: you describe the outcome, test what it delivers, point out what is wrong and request the next change. The cycle is talk, test, talk. The term was coined in 2025 by researcher Andrej Karpathy, half-jokingly, to describe coding by "getting into the vibe" and accepting the AI's suggestions without auditing every line.
Unlike classic no-code (dragging visual blocks in a platform), in vibe coding real code gets written. You are just not the one writing it.
What it looks like: from sentence to screen
A real first session:
- You: "I want a page to track trip expenses with my 4 friends. Each one logs what they paid and at the end it shows who owes whom."
- The AI: generates the page, with a form, a list and the split calculation.
- You: "Nice, but I want it to persist if I close the browser. And give each name a color."
- The AI: adjusts. You test, break it, paste the error message into the chat, it fixes it.
In one evening, the expense splitter exists and the barbecue group is using it. Not a single line of code was read.
What you can really build (and where it breaks)
Works very well for: prototypes to validate an idea before investing, internal tools (calculators, forms, simple dashboards), personal automations, and MVPs to show clients or partners. In short: small software, for yourself or a few people, where failure is cheap.
Where it breaks: security (the AI forgets to lock doors, and you do not know which doors exist), scale (the app that works for 5 friends melts with 5 thousand users), maintenance (when the castle of unread code breaks, finding out where is archaeology) and sensitive data (storing IDs and cards in a vibe-coded app is asking for trouble).
Why programmers roll their eyes (and where they are right)
The criticism is not gatekeeping. Professional programming is 20% writing code and 80% making sure it survives the real world: edge cases, attacks, growth, the colleague who will maintain it. Vibe coding delivers the 20% with flair and ignores the 80%, and a non-programmer cannot see what is missing. For disposable software, great. For software carrying money or other people's data, the 80% is still the product.
The mature reading: vibe coding is to programming what the microwave is to cooking. It genuinely solves a huge class of meals, and restaurants did not disappear.
What changes for those who never coded
The main thing changes: the idea's entry barrier is gone. For decades, having an app idea without knowing how to code meant shelving it or paying dearly. Now the cost of testing dropped to one evening of conversation. Most tests will still die (most ideas die, with or without AI), but now they die on merit, not for lack of access.
There is also an educational side effect: many people who start vibe coding end up wanting to understand what the AI is doing, and learn real logic along the way, with a concrete reason.
How to start tonight
- Pick a small pain of your own: the expense splitter, the secret-santa drawer, the soccer roster organizer. Useful and disposable.
- Describe it as you would to a person: what appears on screen, what happens on click, what needs to be saved.
- Ask for one thing at a time: minimal version first, tweaks one by one. Giant requests produce confused answers.
- Paste errors into the chat: something broke? Copy and paste. The AI fixes its own code surprisingly well.
- Golden rule: no passwords, money or third-party data in your debut app.
Knowing how to code still matters, arguably more than ever: someone has to own the 80%. What ended was the excuse not to try. That idea in your drawer now costs, starting today, one conversation.
Frequently asked questions
The practice of creating software by talking to an AI in natural language: you describe what you want, test what it generates, request adjustments and repeat, without writing or auditing the code. The term was coined in 2025 by researcher Andrej Karpathy.
Yes, within limits: prototypes, internal tools, personal automations and MVPs work well. Applications with many users, sensitive data or money involved still require professional engineering for security, scale and maintenance.
In no-code you assemble software by dragging visual blocks inside a platform. In vibe coding real code is written, just by the AI, from your natural-language instructions, which gives more flexibility and also more ways to break.
The main ones: security flaws invisible to non-programmers, apps that cannot handle user growth, maintenance of code nobody has read, and exposure of sensitive data. Hence the rule: no passwords, money or third-party data in debut apps.
Pick a small personal pain (an expense splitter, a roster organizer), describe the outcome as you would to a person, ask for a minimal version and iterate one request at a time, pasting any error into the conversation.
Data and AI executive with 20+ years building technology that moves businesses. Microsoft Certified Trainer, with executive education at MIT Sloan. At Data Lover, he trains professionals and leads enterprise AI projects.
See profile and all articles →
