Vibe coding is a way of building software where you describe what you want in plain language, an AI model writes the code, and you steer by trying the result and asking for changes. The term was coined by Andrej Karpathy in a February 2025 post, where he described giving in to the flow and not focusing on the code itself.
In practice it looks like a conversation. You ask an AI tool such as Claude Code, Cursor or a browser-based app builder for a feature, run what it produces, and describe what is wrong or what to add next. Many people building this way are founders, designers, product managers and students who would not have shipped software before. Plenty of experienced engineers work this way for prototypes and side projects too.
It is good at getting from an idea to something you can click. Where it gets harder is the stuff that does not show up on screen: security, data handling, performance, and what happens when two features interact. That is where the older habits of software teams help, and none of them require writing code by hand.
A few that are worth picking up early. Use Git, so every working version is saved and you can always go back. Read the diff before you accept a large change, even if you only skim it. Test the thing in a preview before you share it. Ask the model to explain anything you do not understand, including its own code.
Whether you typed the code or prompted it, a pull request still says the same things. That is most of what this dictionary is about.
In the wild
Vibe coded the booking page last night. Before it goes live, could someone review how it stores the customer emails?
