No need, you just ask the AI to "make no mistakes", then you spawn a second AI to review the first AI's output, and then you ask a turdthird AI to review the output of the second AI. Easy peazy. No need for pesky engineers and their stupid human inefficiency. /s
Non-ironically though, I think that's exactly what they did. I have watched Prime's video about this, and, although I was second-screening his video, I believe he mentioned that they used this "technique".