Say what?

Debugging is one of the hardest parts of programming, because any bug might potentially be anywhere in your whole code base, and you need to have a very accurate model of what you expect your program state to be at each step.

No, demos of an AI debugging a one page program that does some well-known operation don't count at all.

I would expect debugging of real-world code to be the very last thing to be automated.