
By Thomas Cohen, founder of Maestro
Vibe coding explained to people who will never code
Describe what you want and let AI code it: vibe coding won over developers in 2025. For a non-developer, one thing is missing: a place to record decisions.
Vibe coding, a term coined by Andrej Karpathy in February 2025 and named word of the year by the Collins dictionary, means describing what you want in natural language and letting AI write the code, judging the result through use rather than rereading the lines. It works for prototyping. It derails as soon as a product must last, for lack of memory.
Why developers love it
A developer who vibe codes keeps two invisible safety nets: they can read the code when something is wrong, and they know the classic traps, lost data, edge cases and security. Vibe coding removes the typing, not the judgement. French articles explaining the practice are almost all aimed at that developer: which editor, which terminal, which shortcuts.
What happens without the safety nets
For a non-developer, the same practice produces a strange experience: the first three requests create a small miracle, then the fourth breaks what the second built. Nobody has done anything wrong; nobody has WRITTEN anything down either. What the product should do exists only in chat history, and an AI asked to ‘add VAT’ does not know it must leave Tuesday's discount decision alone. Without a specification, each conversation starts again from the general mood. That is what vibe means, and it is its limit.
The correction: written decisions
There is no need to learn to code: give the project a memory. A brief stating for whom and why, a requirements document fixing the rules, VAT, Tuesday's discount, and stages checked one by one against those rules. Maestro's method does this work: agents write those documents BEFORE code, you approve them, and every build is checked against your written decisions rather than the conversation's mood. You keep describing things in French, without the drift.
What to remember
Vibe coding proved that description is enough to produce code. Producing a PRODUCT requires more: deciding, writing and approving.