Quality/Agent Debt and Open-Source Resistance

MIT study: 180% more code written, 30% more shipped; GitClear: 81% more duplication, 70% less refactoring. DX Q2 2026: PR sizes doubled, change confidence down 6.1%. GCC and Linux maintainers ban AI-generated code over 15 lines. 1Password study: AI fails to patch 74% of flaws. Vibe coding critiques and 'fake developers' narrative grow. SecureForge reduces security flaws by ~50%. Microsoft's open-source testing agent reduces failures by 63%, addressing trust gap.