@johnnythan That's a good point. 🙂
I don't know how, but they somehow managed it that it usually does not output shitty code.
Maybe rule compliance is better in AI than in humans, like writing and using tests, documentation, etc.
Also, it can just try or look at some stuff much faster than humans can.
Sometimes it is also running a wrong direction though, like a junior developer.
That's why someone experienced has to look at the code and the design.