When the bot breaks the code, is it generally best to let it try to fix, or to revert to a previous version and try something different?
I’ve tried ‘auto-fix’ many times with limited results, which makes sense. If the bot made a mistake, it’s likely not able to see what the problem is, or it should not have made the blunder in the first place.
In this case the auto-fix caused the bot to regenerate practically the entire code base, taking seven minutes to run, and burning a thousand tokens. (But it did work.)
I’ve got into the habit if rolling back when things aren’t going the right direction, but that too gets expensive, and I’m making limited progress.
