We're still producing code the same way since we started in the 70's. We expect stateless LLMs to work with code the same way as very stateful human dev teams.