I have been using Google AI studio but it is error prone. If I get Claude Sonnet to tell me the errors and then tell Google Ai studio the errors then it can fix them quite well. But it may still decide to drop a key part of the code. For example, you might have had data sorted and the sort gets dropped. You may only find out days later because your mind is focused on the other broken thing.

Most days I have to say - I have had to go back to an earlier working version.