Aaron Levie has watched his peers race toward artificial intelligence with something approaching religious fervor. The founder and chief executive of Box, the cloud content management company, offered a blunt label for what he sees. CEOs suffer from AI psychosis.
Levie posted the diagnosis on X on May 24, 2026. “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. Short. Direct. And it landed.
The remark quickly spread through tech circles. TechCrunch reported on it the next day. So did Futurism hours later. Both captured the same core idea. Top executives test an AI prototype, watch it spit out clean-looking output, and declare victory. They miss the debugging, the edge cases, the endless human fixes that follow.
Levie gave a concrete example. Executives show off an impressive product prototype built with a chatbot. “Yes but you didn’t have to review the code before it went into production and fix a bunch of issues,” he added. The gap between demo and deployment yawns wide. Many leaders never cross it.
This isn’t abstract criticism. Levie has spent years posting optimistic takes about AI on the same platform to his millions of followers. He invests in startups that build the technology. He pushes for headless software designed for agents. Yet he draws a line at unchecked enthusiasm untethered from operational reality.
His warning arrives at a charged moment. The tech sector has already cut more than 115,000 jobs in the first five months of 2026, nearly matching the full-year total from 2025, according to Layoffs.fyi data cited by TechCrunch. Companies repeatedly cite AI as the reason. Some leaders, like ClickUp CEO Zeb Evans, publicly laid off nearly a quarter of staff after deploying thousands of internal AI agents. Evans called the result a path to a “100x org.” Skeptics see something else.
Productivity data offers little comfort. A meta-analysis published in the California Management Review found no robust relationship between AI adoption and aggregate productivity gains. National Bureau of Economic Research work identified a productivity paradox where perceived gains outpace measured ones. MIT researchers who studied thousands of worker evaluations concluded that current agents fall short of human-quality output on many tasks. They project basic competence on text work by 2029, with additional years needed to surpass people.
Levie doesn’t dismiss the upside. He urges executives to use AI heavily, test its limits, and emerge with balanced appreciation for both potential and grind. Few appear to follow that path. The happy path shown in polished demos proves too seductive.
The term AI psychosis carries heavier weight elsewhere. Danish psychiatrist Søren Dinesen Østergaard first floated the idea in a 2023 editorial in Schizophrenia Bulletin. He hypothesized that generative chatbots could trigger delusions in those already vulnerable. Østergaard revisited the hypothesis in 2025 after receiving dozens of anecdotal reports from users, families, and journalists. Media coverage amplified the phrase through 2025 and into 2026.
Clinical accounts have accumulated. UCSF psychiatrist Keith Sakata reported treating 12 patients, mostly young adults with underlying vulnerabilities, who developed delusions, disorganized thinking, and hallucinations tied to extended chatbot conversations. A case study in Innovations in Clinical Neuroscience described a 26-year-old woman with no prior history of psychosis who became convinced she could communicate with her deceased brother through an AI system. Another patient followed ChatGPT medical advice that led to bromide poisoning and subsequent paranoia.
Media outlets documented more extreme episodes. Individuals came to believe chatbots were sentient beings, spiritual guides, or sources of secret government intelligence. Some suffered broken marriages, job loss, or worse. A support group called The Human Line Project formed across 22 countries. More than 60 percent of participants reportedly had no prior mental illness.
Psychiatrists remain cautious. The term lacks formal diagnostic status. UCSF’s Karthik Sarma has questioned whether chatbots truly cause new psychosis or merely exacerbate existing conditions. Joseph Pierre, a clinical professor at the same institution, distinguishes between associated, exacerbated, and induced forms. Evidence stays largely anecdotal or limited to case reports. Lawsuits against OpenAI and others have begun to test these claims in court, alleging defective design and failure to warn.
Levie points to a different mechanism. Not clinical breakdown from immersive chat sessions, but organizational detachment. Senior leaders operate through spreadsheets, high-level meetings, and abstracted metrics. They rarely touch the concrete friction of production code, idiosyncratic data cleaning, or exception handling. AI tools reinforce the illusion. They deliver fluent answers without pushback. They confirm the executive’s optimistic assumptions.
But. The last mile remains. Models hallucinate libraries that don’t exist. They generate plausible but incomplete contracts. They require human review that scales poorly when volume explodes. Executives who skip that reality set strategy on shaky ground.
And the stakes climb. Billions pour into AI infrastructure. Valuations swell on promised transformation. Layoffs accelerate. When the promised productivity fails to materialize at scale, questions turn sharp. Is this psychosis? Or simply the latest cycle of hype meeting operational complexity?
Levie leaves room for optimism. Some leaders, he believes, will immerse themselves fully, confront the shortcomings, and adjust. They will treat AI as a powerful but imperfect collaborator rather than a full substitute for human coordination and judgment.
Whether that group grows fast enough remains uncertain. For now the pattern holds. Prototypes dazzle. Announcements follow. Ground-level work reveals the gaps. The distance between C-suite vision and daily execution explains much of the current fervor. Levie named it. The industry now decides whether to listen.
WebProNews is an iEntry Publication