There's a peculiar energy running through the tech industry right now—one that echoes past upheavals like the early days of cloud computing, when costs spiraled out of control, yet also feels entirely unprecedented, with record revenues coinciding with mass layoffs. A circulating theory tries to make sense of this paradox: tech executives, particularly CEOs, are collectively suffering from delusions of grandeur fueled by artificial intelligence. And at least one tech leader has said it out loud: Aaron Levie, founder of Box.

"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," Levie posted on X. He argues that CEOs "play with AI," develop a prototype, or generate a contract—using his own examples—and then leap to the conclusion that AI agents can handle the real work. But these top executives aren't the ones reviewing code, uncovering bugs, or catching calls to hallucinated libraries before software ships. They aren't responsible for training AI models on a company's unique contract terms, nor do they spend days combing through agreements to find hidden clauses, as Levie points out. In short, his theory suggests that CEOs lack a deep enough understanding of processes to know what can truly be automated—yet that ignorance doesn't stop them from acting on their beliefs.

It's worth noting that Levie isn't an AI skeptic. Quite the opposite. He frequently posts AI-positive content on X to his 2.7 million followers, writing blog posts like "Headless software is the future" about how software built for AI agents is the way forward. He also backs up his words with action, actively investing in AI startups as an angel investor. So what does Levie recommend CEOs do instead? He advises them to use AI "a ton" to truly grasp its capabilities and limitations, "and come out the other side with an appreciation for both the upside and the real work."

I have enough faith in humanity to believe some CEOs are genuinely trying to do that, but right now, they appear to be in the minority. In just the first five months of 2026, the tech industry has already seen nearly as many layoffs as in all of 2025: 115,430 people have been let go from 152 tech companies so far this year, compared to 124,636 people laid off by 275 companies in 2025, according to industry layoff tracker Layoffs.fyi. And most companies have cited AI as a reason for these cuts. Many argue that the biggest tech firms are simply "AI washing"—crediting AI for productivity gains, past or future, when other business decisions and metrics are really driving the reductions. Still, some stories are particularly striking.

Zeb Evans, CEO of project management and productivity software startup ClickUp, proudly announced on X that he had laid off nearly a quarter of his workforce—22%—after deploying about 3,000 AI agents to handle internal tasks. Evans insisted this wasn't about cutting costs. Instead, he envisions a workforce composed of people who manage AI agents and spend their days quickly reviewing the agents' output. He believes this will create what he calls a "100x org." While AI can be a highly useful tool, the data on AI and productivity doesn't support such assumptions—by a wide margin. A meta-analysis of other research, published in October in UC Berkeley's California Management Review, found "no robust relationship between AI adoption and aggregate productivity gain."

Research published in March by the National Bureau of Economic Research did conclude that AI adoption improves productivity, but it also noted "a productivity paradox, in which perceived productivity gains are larger than measured productivity gains." After creating thousands of agents to work on tasks, researchers at MIT concluded that agents simply aren't producing human-quality work yet in many cases. They predict that, at the current rate of LLM improvement, models will "be able to complete most text-related tasks with success rates of, on average, 80%-95% by 2029 at a minimally sufficient quality level." In other words, AI is on track to achieve basic competence on most tasks in about three years. These researchers believe agents will need a few more years to outperform humans.

Meanwhile, research published in the Harvard Business Review showed that when everyone uses AI to produce more output, the bottleneck simply shifts to executives. Their work now awaits the people who must authorize all the material being generated. If everyone is empowered to act, then—based on what OpenAI experienced last year—things may spiral out of control. Are CEOs ready for that? If not, the most likely outcome of the ongoing CEO AI psychosis is simply organizational chaos.