That future version of Claude will still be mediated by an error prone human who in this case would be a non-engineer. So even if Claude is not error prone, it could be giving correct answers to the wrong questions and correct designs for the wrong constraints because the human getting Claude to do things doesn't know any better.