The Turing Test didn't anticipate super-parrots that converse nicely, instead assuming that an AI would actually reason to participate in a conversation.

The "AI" we have now isn't actually "I".

and yet it fails to fix any real bug e2e in a large enough codebase. It require a lot of babysitting to the point that actual performance boost is very questionable