It's very fustrating when you end up spending more time tracking down bugs and cleaning up inconsistent code then it would have taken to do it right yourself in

It's very fustrating when you end up spending more time tracking down bugs and cleaning up inconsistent code then it would have taken to do it right yourself in the first place. It's an important lesson. That said, there are still massive productivity gains available. If you let the model mess with your existing code base, or if it get's past a small size, it will shit all over it. The LLM needs limited scope. I find it's great for frontend work, which is what I hate doing myself. I've been having it build webcomponents, paired with targeted backend functions that wrap internal libraries that I wrote for db and API calls. Then I compose things. Careful instructions, internal documentaion for the model and well contained isolated tasks are best. The models get confused and distracted by existing code, it's ofen better to document function interfaces and use rather than showing it the source. If the code is going into production you have to read it line by line and clean it up. I've been using Aider, mostly.