Human experts’ time is precious and limited, while LLMs are scalable but can make mistakes. Do we really have to choose between them, or is there a better way to allocate work across both? Teams building AI systems often compare two candidate answers to see whether a change made the system better.