i agree with that but now at the pace where everything is moving really really fast, humans are using ai tools n all for imensively writing code and building new feature where they are also facing issues and frustration while debugging...llms are really good but sometimes they catch the issues but majority of time, they just grep or bash the things and eventually ending up not losing context....but if we go with your proposed way, then if humans build n llms review then it will be the slow process.....Hope I put my point correctly