Nevertheless, the problem of determining what plugins it needs is actually harder than writing these plugins itself.
That is very arguable.
You can have unit tests, speed execution tests and all kind of tests to see if a plugin is good or not.
A program writing programs? I think we don't have that yet (well, there is
https://en.wikipedia.org/wiki/Evolutionary_algorithm but from what I read in another thread DAC-related, it takes huge computational power).
Basically this. Your core modules will have to have near perfect, and have the best unit tests. From there its just evolution and experimentation.
The solution is to built from solid parts.
Some people think this is highly complex. It is, but not unobtainable. The fundamental of computer science is take one big problem you can't solve, and split it into many small problems that you can solve.