If you don't even have a compiler, it's easier to patch a binary that's already running than make one from scratch. That's a closer analogy to the situation under discussion.
There are other ways of doing this even without germline editing - you could for example train a linear transformer model that does DNA phenotype prediction, then just do regular IVF and choose the embryo with lowest predicted risk of (cancer, heart disease, etc)
The only roadblock to doing this right now is access to training data.