You're Not Tired From Using AI. You're Tired From Checking It.

New HBR research on "AI brain fry" is worth five minutes of your time — but read it knowing who wrote it.

The finding that actually holds up: AI fatigue isn't coming from doing more work. It's coming from managing the work AI produces. The study separates the two cleanly, and most coverage misses that distinction. There's a difference between a heavier workload and a different kind of cognitive load — and the paper identifies the second thing with some precision. The 33% spike in decision fatigue among workers supervising multiple AI outputs isn't a burnout story. It's a workflow design story.

The part that doesn't hold up is the conclusion. The authors — all from Boston Consulting Group, which advises enterprises on AI adoption — land on "you're using it wrong" as the diagnosis. High-oversight patterns are the problem; better delegation is the fix. That's a convenient answer if your business model involves selling organisations better AI implementation.

What the paper never asks: why are workers in high-oversight mode in the first place? Because the tools produce errors at a rate that makes checking rational. Treating the verification instinct as a fatigue management failure skips the more honest question of whether the oversight burden is proportionate to the actual reliability of the outputs. The humans aren't misconfigured. They're responding sensibly to tools that aren't yet trustworthy enough to stop watching.

The one number to save: workers experiencing this cognitive load are 10% more likely to quit. That's not a wellness metric. It's an unbooked deployment cost, and it belongs in every AI ROI calculation that currently leaves it out.