Well, if you had tried using GPT/Codex for development you would know that the output from those 5 hours would not be 90% trash, it would be close to 100% pure magic. I'm not kidding. It's incredible as long as you use a proper analyze-plan-implement-test-document process.