It’s also often possible to know what a human will be bad at before they start. This allows you to delegate tasks better or vary the level of pre-work you do before getting started. This is pretty unpredictable with LLMs still.