AI is going to have to get a lot better before "Programmed by a human" is an anachronism. I played around with Copilot over the weekend because everyone on HN urged me to try it and thought it would change my mind about AI. It's pretty terrible still. I tested three things on a few existing C++ and Objective-C code bases: 1. Have it explain existing code to me: This worked reasonably well. I give it an A-minus. AI seems to be able to analyze and explain code at the level of a junior CS student in college. 2. Have it write new code to do a few tasks: I give it a C grade. Not bad, but a lot of the code it writes just isn't how I'd do it, and I saw a couple of subtle bugs. 3. Suggest improvements/refactors to existing code: I give it an F. It's terrible, and often provides suggestions that don't make sense (the return value for foo() should be bar, when it already is bar!) or that will actually make the code worse.

I think it will be a long time before AI produces code of such quality and correctness that I'd give it a passing grade in a college-level CS/programming course.

One thing I give it credit for, it will confidently spit out code when you ask it, often code that works, but it's pretty terrible code.