This is how I’ve been building charcoal.
I spent a lot of time thinking about *how* the architecture of the application would go together, and reading papers on the concepts I didn’t understand. I have also relied on tooling built by professional software developers to help constrain model slop.
in my experience there is now room to “vibecode” a prototype, evolve it into a more serious app, and then down the road refactor it into something that is more professional, organized, and sustainable
previous AI models couldn’t handle large refactors like this, but they now can with human aide