Harvard Business Review editorial · Harvard Business Review · 2026-05-16
The February 2026 HBR piece names a paradox the productivity-study literature has been circling around: workers using AI report doing more work, not less, and the work they do is more cognitively demanding, not less. The labour-saving frame is wrong; "labour-intensifying" is closer.
The mechanism. AI removes the easy parts of knowledge work — first drafts, lookups, summary generation, boilerplate. What's left is the parts that require judgment: editing, verifying, deciding, integrating, taking responsibility. Output goes up because the easy parts no longer rate-limit you. Cognitive load goes up because everything you do now is the hard part.
The wellbeing tax. HBR pairs the productivity story with what they call "oversight fatigue": the mental cost of reviewing AI outputs you cannot fully trust. Verification is itself work, and at sufficient volume it dominates the time saved by generation. The associated Fortune piece on BCG's "AI brain fry" study finds heavy AI users reporting higher rates of decision fatigue, weaker recall of their own work, and lower confidence in their judgment — the precise opposite of the "AI as cognitive prosthesis" pitch.
The UC Berkeley counter-evidence. The HBR piece flags Berkeley 2026 work showing the intensification effect is not uniform — workers with high autonomy over how AI is integrated into their workflow report neutral or positive wellbeing effects, while workers under top-down AI mandates report the worst outcomes. The variable is control, not exposure.
What it means for AI literacy. The implication is that AI training programs focused only on capability miss half the problem. The other half is workflow design: when to use AI, when not to, how to structure verification so it doesn't dominate the time saved, and how to preserve the cognitive skills the AI is fluent in so you don't lose them to disuse.