The NVIDIA report highlights engineers retaining accountability over AI recommendations, but I'm still wondering if 'room to question' exists when the workflow design itself limits that right.