Superhuman AI coding still makes software worse

AI can write code faster than humans, but speed without quality creates a compounding defect problem. If AI submits 10x more changes, even at a lower defect rate, total bugs still multiply. Add model collapse in AI-heavy codebases and growing context limits, and you have a recipe for technical debt at scale. Quality-first thinking and human oversight remain essential.