Right now AI is great at plumbing — CRUD endpoints, test scaffolding, repetitive config. But the hard decisions — what to build, how services should talk, what to NOT build — still feel very human. I wonder how long that distinction holds.