Is there a blocker that would prevent future AI to write perfect assembler (for n architectures) in 1st pass?