Few words have spread as fast in the AI world as "vibecoding". In less than a year it went from an offhand remark to one of the industry's most used terms. But what does it actually mean – and why is everyone talking about it?
Where the term came from
The expression was coined in early 2025 by Andrej Karpathy, one of the more influential voices in AI. He described a way of building software where you "give in to the vibes" – letting the AI write the code while you steer with words and intuition, rather than reading every line.
It captured something many already felt: that the bar for building something had shifted dramatically.
What it actually means
In practice, vibecoding means you describe what you want in plain language, the AI generates the code, and you iterate by reacting to the result – "make the button bigger", "add login" – instead of writing it all yourself.
The effect is that more people can build. Someone with an idea but no developer background can get further on their own than ever before. It's part of why the whole field around AI – and the role of AI consultant – is so new and growing so fast.
Where it fits – and where the line is
Vibecoding is hard to beat for exploring ideas, building prototypes and internal tools quickly. Where speed is everything and the margin for error is wide, it's enormously useful.
The line is drawn at what has to live in production – run every day, on real data, with real consequences. There the vibes aren't enough: it takes judgement about security, maintenance and what actually holds. The technology makes the start faster; the judgement about what to harden is still the work.