Not only does it fix it, it apologizes for fukking up, tell you where it fukked up at, explain what it did to fix it and rewrite the code. Claude even helped me set up supabase. This shyt is wild as fukk. At one point I was still chugging along learning JavaScript and I just said fukk it. I didn't feel like learning the complexities of React. I just starting learning to use prompts and planning out what I wanted to build. The only thing these models can't do yet is pull from more libraries. But in about another year they probably will.
Time management is the biggest thing though. I try not to waste questions, because it eats up tokens. I wasted a lot of times asking it to add or change shyt because I went in blind. Now I use ChatGTP to actually map out what I want organize the project into phases and just build from there. If I want to do a custom design, I will fine something that's already been done copy the code, feed to to Claude and it will build it for me.
I would say the most work I do, is copying all the code in my files to txt docs. One of the setbacks with Claude is when you have long chats, it eats up your useage. So before I end the chat, I ask it to write a summary of what was done, and use that to start a new chat. Cursor is dope too, but I don't really use it that much. But these models are only $20 a month. The only AI that I haven't paid for is midjourney. It use to be free but they started charging for it because so many people are using it.
can you make an example of your process and share the chat please?