AI tools are increasing velocity at the developer layer but that is adding pressure on longstanding bottlenecks in the management layer, with R&D leaders now trying to manage a high-speed, AI-powered development engine using the same manual tools and processes. What needs to change?