Building a web framework with AI while refusing to lower the bar on code quality

A developer built Tango, a TypeScript meta-framework inspired by Django and Django REST Framework, using AI coding tools while enforcing strict code quality standards. The workflow involved TDD with 100% coverage, adversarial multi-agent design reviews, agent 'skills' to persist learned lessons, and dogfooding via fresh agents given only documentation. Key AI failure modes encountered include preference for locally correct solutions over holistic design, insertion of compatibility shims during refactors, and sycophantic behavior. The author shares reusable agent skills in the GitHub repo and reflects on AI as a powerful but constrained tool rather than a trusted collaborator.