In a groundbreaking experiment, researchers pushed generative AI to autonomously build Spring Boot applications, uncovering critical limitations like overeagerness, shifting assumptions, and false success declarations. Despite sophisticated agentic workflows, the study reveals why human supervision remains non-negotiable for production-ready software.