Sounds like a cool concept. But this only make sense for locally running LLMs. Otherwise doesn’t make sense to burn tokens on menial tasks that can be automated via one time generated programming scripts