If AI writes the code, who tests it?

I watched a developer ship a complete booking feature last Tuesday. Start to finish, about three hours. The kind of thing that would have taken a sprint two years ago. He'd used Cursor, Claude, some Copilot autocomplete. The feature worked in the demo. It crashed in staging within fifteen minutes because nobody had tested what happens when two people book the same slot at the same time. That moment keeps replaying in my head, because it captures the exact problem nobody in the AI hype cycle wants to talk about. We've gotten incredibly fast at producing code. We have not gotten any faster at knowing whether the code actually works. The speed trap Here's what's happening in practice, not theory. Tools like Cursor, Copilot, and Claude Code let developers produce features at a pace that was unthinkable three years ago. A solo developer can now vibe-code an entire CRUD app in an afternoon. Teams are shipping features weekly that used to take months. That's real and it's genuinely impressive