A new term is making the rounds in tech leadership circles, and it’s not flattering: “AI psychosis.” According to TechCrunch AI, Box founder Aaron Levie coined it to describe a specific kind of executive delusion, and the diagnosis is landing because the symptoms are everywhere. Record revenues paired with mass layoffs. Billion-dollar agent rollouts. CEOs swearing the robots are ready while the people doing the actual work know better.
What stands out here is who’s making the argument. Levie isn’t an AI skeptic. He posts relentless AI optimism to his 2.7 million followers on X, writes blogs like “Headless software is the future,” and backs AI startups as an angel investor. So when he says CEOs are losing the plot, it carries weight.
The core of Levie’s theory
Levie’s point is about distance. “CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI,” he wrote. Executives play with a chatbot, generate a prototype or a contract, and make the leap to believing agents can run the whole operation.
But they’re not the ones reviewing code, catching bugs, or spotting calls to hallucinated libraries before deployment. They’re not training models on a company’s weird, specific contract terms. They don’t understand the process well enough to know what can and can’t be automated. That gap in knowledge doesn’t stop them from acting. That’s the dangerous part.
The layoff numbers tell the story
The behavior shows up in the data. As TechCrunch AI reports, the tech industry has already racked up nearly as many layoffs in the first five months of 2026 as in all of 2025. The tally: 115,430 people cut from 152 companies this year, versus 124,636 from 275 companies last year, per Layoffs.fyi. Most companies blamed AI.
Many of those claims are what critics call “AI washing,” crediting AI for cuts that other business pressures actually drove. But some executives are true believers. ClickUp CEO Zeb Evans publicly announced he’d laid off 22% of staff after deploying roughly 3,000 internal AI agents. He insisted it wasn’t about cost. He wants a workforce that runs agents and reviews their output, building what he calls a “100x org.”
What the research actually says
Here’s where the believers run into trouble. The evidence doesn’t back the hype:
- A meta-analysis in UC Berkeley’s California Management Review (October) found “no robust relationship between AI adoption and aggregate productivity gain.”
- A March National Bureau of Economic Research paper found real gains, but flagged a “productivity paradox” where perceived gains outrun measured ones.
- MIT researchers who built thousands of agents concluded the agents aren’t doing human-quality work in many cases yet. Their forecast: models hitting 80–95% success on most text tasks by 2029, with a few more years after that to actually beat humans.
Translation: AI is on track for base competence on most tasks in about three years, not today.
The bottleneck nobody’s planning for
There’s a second-order problem most CEOs miss. Harvard Business Review research found that when everyone uses AI to produce more, the bottleneck shifts upward. All that extra output needs someone to authorize it, and that someone is usually an executive. Empower everyone to act, and you get the kind of runaway dynamics OpenAI wrestled with last year.
So the question Levie’s theory raises is sharp: if you fire a quarter of your staff and unleash thousands of agents, who’s catching the errors, and who’s approving the flood of work? If the answer is “nobody planned for that,” the result isn’t a 100x org. It’s organizational chaos.
The practical takeaway
Levie’s advice is the right starting point, and it applies beyond the C-suite. Use AI “a ton” before betting the org chart on it. Sit with the real work long enough to learn what it genuinely automates and where it quietly fails.
For practitioners and operators, the move is to document where agents break, keep humans on review for anything that ships, and treat productivity claims as hypotheses to test, not facts to act on. The CEOs who survive this stretch won’t be the loudest believers. They’ll be the ones who did the homework. You can read the full breakdown at TechCrunch AI.