i had idea like this when gpt first came out but decided not to go down this route because idk at what point to automate and at what point to refine and just seemed too complex and now tools like claude cli seem to make more sense for devs. but as internal b2b tool for non technical with predefined schemas and apis and very strong ui rules (eg u must use this library/component/pattern at all times for X or Y) u COULD generalize but then it would also be impotent to fork and extend those that already been made and without bugs or be able to fix bugs..and the second u need something slightly non trivial would need to be able to “drill down” to fix this or fix that or tweak or whatever so just becomes to vibe code and deploy so maybe like to vibe code harder and faster and deploy faster but maybe with enough rhfl it would be viable… i can imagine like some sort of fluid experience especially for learning or research where u just need to process and visualize some data and don’t care about stuff breaking because its internal tool