Box's Aaron Levie: CEOs Have 'AI Psychosis' Because They're Too Far From Real Work

Box CEO Aaron Levie has coined the term ‘AI psychosis’ to describe a pattern he sees among tech executives: they prototype something with AI, get impressed, and leap to the conclusion that agents can replace entire workflows. Levie’s argument is that CEOs sit too far from the last mile of work — debugging hallucinated library calls, training models on quirky contract terms, catching subtle errors — to grasp what genuinely automates and what doesn’t. He’s not an AI skeptic; he’s an active AI investor urging peers to actually use the tools heavily before betting headcount on them.

The stakes are visible in the layoff numbers. In the first five months of 2026, 152 tech companies cut roughly 115,000 jobs, nearly matching all of 2025, with AI cited as the rationale. ClickUp CEO Zeb Evans publicly celebrated firing 22% of staff after deploying around 3,000 internal agents, pitching a ‘100x org’ of humans who review agent output rather than produce it themselves.

The research doesn’t back the bet. A UC Berkeley meta-analysis found no robust link between AI adoption and aggregate productivity. NBER flagged a gap between perceived and measured gains. MIT projects agents won’t hit 80–95% success on most text tasks until around 2029, with human-beating performance further out. A Harvard Business Review study notes that when AI accelerates output across an org, the bottleneck shifts to executive approval — meaning the same CEOs driving the cuts may become the constraint they failed to anticipate.