Hundreds of firms show that human code review creates a critical bottleneck for AI tools. While AI agents generate functional code rapidly, the effort required to verify accuracy cancels out these gains. This friction prevents firms from increasing total software output or reducing staff.