However, someone will have to know how to tweak the genetic algorithms and feed them the right data. If you think it will ever be "just specify the problem and run the GA/GP", you are gravely mistaken. It is not that simple, and there will be ways of phrasing the problem that will make it easy for the GA and ways that are hard, and problems that GA will not be able to deal with at all. There will still be programmers.

As I recently said in another post, the gulf between GP hype and GP reality is pretty much the largest I know of in the realm of computer science. There's a lot of very tricky theory, and my expectation is that this will never be a practical way to program (as opposed to fixing bugs), and even if it is, it will take a lot more knowledge and skill than a current programming job, which also means that fewer people will be able to do it. In my estimation, it will just plain be harder than writing it conventionally and thus never be practical.

So you could end up with fully-test driven development (requirements oriented?).

I kind of doubt genetic programming is a way to fully automated development.

Someone will always have to describe what it is that the computer should do.