The last time I asked a AI for code was for a code snippet to do orthographic transcription (e.g. Владимир Путин -> Vladimir Putin). The AI responded with a snippet doing a (even pointless) change of encoding that would result in an exception only.
I complained about getting only an exception when executing the code. It then try catch print stack trace the code. After complaining about not wanting a stack trace, the AI removed the stack trace output, so the result was just nothing.
So I complained again and reformulated my requirements. It then corrected the code to a version that did again only change the encoding transitionally resulting in strings containing replacement characters for the ones not available in the target encoding.
I gave up. 🤦‍♂️ 

What I did expect was giving me a solution using some library or telling me a common algorithm or library for orthographic transcription is not available.
This is just one example. Almost always when I ask a AI for non trivial code the response is just wrong, fantasizing about non existing APIs or contains computational errors. Corrections after rejecting responses are so stupid that if I did get from team mates I would have asked them if they want to be kidding me.