Non-coders can ship real apps with AI — until, suddenly, they cannot.
The tools are genuinely new. A person with an idea and no coding background can now produce a working application in a week. The speed is real. The stall is also real. One morning they open the codebase and realise they have built something they cannot read, cannot debug, and cannot extend.
The pattern repeats in predictable ways: a 70% wall where context collapses and debugging becomes impossible; a dependency trap where technical debt accumulates invisibly; a prompting paradox, where you cannot ask for what you do not know exists. The honeymoon ends, and the builder is left with a codebase that runs but does not belong to them.
The two available answers both fail. A traditional bootcamp is too slow and aimed at the wrong outcome. Pure vibe coding ignores fundamentals and produces apps that cannot be maintained. There is a third path.
A mental model, not a syllabus.
VibeMastery is organised around one bet: builders do not need to learn programming in the classical sense, but they do need to understand how their application actually works. Enough model to read AI output critically, prompt with precision, and debug with context.
- Progressive milestones — each one builds a single layer of the mental model. What happens when you click a link, then frontend, backend, servers, databases, frameworks, Laravel, and the vocabulary needed to read an AI-generated plan and push back on it.
- Project-centred, not curriculum-centred — the work is always a real application the builder is trying to ship, not a toy exercise that forgets itself the moment the lesson ends.
- Laravel and Claude Code as the stack — an opinionated pair, chosen because they are legible, well-documented, and play unusually well with an agent that writes code alongside a human who is learning to read it.
- Plain language — long-form articles, short posts, and video scripts, written without jargon for its own sake, with analogies that do actual work.
The hardest part is not the teaching. It is the restraint. Every lesson has to earn its place against the builder's own momentum — the course competes with the thing they are trying to ship, and it has to make that thing better, not slower.
Five leads, all of them developers. Sunset.
The thesis held up. The market I drew in did not. Every one of the first leads was a working developer looking for sharper AI workflows — a real audience, but not this audience. The product was built for idea people and wall-hitters, and the language that made it legible to developers was the same language that kept non-coders from finding it.
Rather than retarget the course into something it was never meant to be, I sunset it. The course is closed. The community is closed. The domain stays up as a redirect.
A thesis can be right and still not find its people.
The lesson is not that the positioning was wrong. It is that a positioning written by a developer, published through developer channels, read by a developer audience, will quietly collect developers no matter how many times the copy says non-coder. Distribution is part of the product, and I built the product before I built the distribution.
The writing, though, was the best thing about the project. The milestones on mental models — what happens when you click a link, how a request reaches a database, why an AI-written plan is only as good as the vocabulary you can push back with — those pieces work on their own. They are moving to my writing, where they do not have to carry a business on their back. I will keep posting about the same questions there.
The course is done. The argument is not.