If one AI could clean other AI mistakes, shouldn't the first said be able to not make such mistakes?

AI isn't AI. It's predictive text. They will all fall in the same pitfalls as any other. "AI" can write code and can be very helpful making short functions, but it can't and will never be able to do or work in whole systems because, just by nature, those are made to fit human needs, illogical and sometimes very specific needs that can't be just "predicted".