All the failures and faults become training data. Even this conversation. Training data. It will “just work” eventually, or OpenAI will host vibe coded projects that are “ready to use” in a few seconds. Like a GitHub or docker registry where the admins are all LLMs satisfying the world’s software requirements.