10

Brain Fry

The supervisor tax

Sustained engagement with AI produces cognitive fatigue, driven not only by the volume of AI use, but by the load of monitoring, evaluating, and correcting AI output across multiple concurrent workstreams.

A BCG/HBR study of 1,488 workers (March 2026) put a number on it: practitioners who supervise AI outputs report 12% more mental fatigue, 14% more cognitive load, and 19% more information overload than those who don't. The mechanism is specific. Drafting with AI is manageable. Verifying AI output requires sustained vigilance against subtly wrong but confident-sounding material, and that review depletes cognitive resources faster than almost any other mental activity. This is the supervisor tax: the hidden cost of augmentation that the productivity narrative ignores. Three concurrent AI workstreams is the maximum sustainable monitoring load for most people; beyond that, output quality drops. Human productivity is not infinitely scalable. And computers do not share any cognitive value system with us.

Design transfer

  • Output quality degrading across the day as monitoring load compounds
  • Decision-making capacity eroded for the high-judgment work AI cannot do, because the cognitive budget was spent on verification
  • Reduced capacity for the deep, unstructured thinking design relies on, replaced by reactive oversight
  • AI fatigue framed by leadership as an adoption failure rather than a load-management signal

In the wild