In many cases it's also still better to write code by hand as reviewing is not as good as writing the code yourself, as not being versed in the code makes that your review takes quite a bit more time than earlier. This as you have to comprehend what the model did. Also, the models have been trained on public code and with the amount of below-average codebases on GitHub, especially for languages as JavaScript and Python, AI has the tendency to implement quite a bit of technical debt and over complicating things.
It's a bit like the autocorrect on your phone while swiping. It makes you write faster, but usually you have to do quite a bit of fixing afterwards as the autocorrect always wants to write silly words. And quite often, you miss things, leaving some silly words in your message.
