Vibe Coding: Building Software by Describing It

In February 2025, Andrej Karpathy, a founding member of OpenAI, posted a short note that named something already happening everywhere: 'There's a new kind of coding I call vibe coding, where you fully give in to the vibes... and forget that the code even exists.' Instead of typing every line, you describe what you want in plain English, an AI writes the code, you run it, and if it breaks you just say 'fix that' and keep going. The phrase spread so fast it entered the Merriam-Webster dictionary within months.

This is a genuine paradigm shift, a change in the basic rules of how work gets done. For fifty years, programming meant memorizing exact syntax: a missing semicolon could break everything. Vibe coding moves the human's job up a level, from writing syntax to describing intent and judging results. This lesson looks at what actually changes, what stays the same, and where the traps are.

- Define vibe coding and explain what shifted about the programmer's role. - Distinguish tasks where vibe coding shines from tasks where it's risky. - Explain why clear description and testing matter more, not less, in this model. - Practice writing a precise prompt that would produce working software.

The classic model treated the human as the author of every instruction. The AI-assisted model treats the human as a director. You still decide what to build and whether the result is good, but the AI drafts the code. Karpathy described building small apps this way, accepting changes without reading every diff, and leaning on the AI to debug its own errors. The skill that matters most becomes the ability to describe precisely what you want and to notice when the result is wrong.

Notice the shift is in altitude, not abandonment. A director who cannot tell a good scene from a bad one makes a bad film. A vibe coder who cannot test whether the app actually works ships broken software. The judgment moves up; it does not disappear.

Vibe coding is excellent for prototypes, personal tools, learning by experimenting, and getting unstuck when you don't know the syntax. A student can build a working quiz app in an afternoon without a computer-science degree. That is a real democratization of who gets to build.

It bites when stakes are high and details are subtle: handling money, storing passwords, medical or legal logic, or code that must scale to millions of users. In those cases you still need to read, understand, and rigorously test the code. The best practitioners in 2025 treated the AI as a fast junior developer whose work they always reviewed, not as an infallible oracle.