Tech CEOs Are Reportedly Falling Into “AI Psychosis”

The tech industry is currently experiencing rapid, sometimes chaotic change driven by artificial intelligence, with some executives making increasingly aggressive claims about AI’s ability to replace human work. Critics argue that many tech leaders may be overestimating what AI systems can realistically accomplish today.

One of the more striking descriptions of this trend came from Box founder and CEO Aaron Levie, who recently referred to the phenomenon as “AI psychosis” among executives.

In a post on X, Levie argued that CEOs are especially vulnerable to unrealistic expectations around AI because they are often removed from the day-to-day operational work required to make AI systems function effectively in real business environments.

According to Levie, many executives experiment with AI tools by generating simple prototypes, drafting contracts, or automating small tasks, then incorrectly assume that AI agents can fully replace complex workflows handled by employees.

However, the reality is often far more complicated. Engineers still need to review AI-generated code, identify bugs, and correct hallucinated outputs before software can safely be deployed. Legal teams must still manually examine contracts, train AI systems on company-specific policies, and verify sensitive details that AI tools may miss or misunderstand.

Levie emphasised that the issue is not a rejection of AI itself. In fact, he remains one of the technology industry’s more vocal supporters of artificial intelligence and regularly promotes AI startups and AI-driven software development. He argues instead that executives need deeper practical experience with AI before making major operational decisions based on overly optimistic assumptions.

The discussion comes at a time when layoffs across the technology sector continue accelerating. According to industry tracker Layoffs.fyi, more than 115,000 tech employees have already lost their jobs across 152 companies during the first five months of 2026 alone — nearly matching the total layoffs recorded throughout all of 2025.

Many companies have directly linked workforce reductions to AI adoption and automation initiatives. Critics, however, argue that some firms may be overstating AI’s actual impact while using it to justify broader cost-cutting measures.

Project management software company ClickUp became one of the more controversial examples after CEO Zeb Evans said the company had reduced its workforce by roughly 22% following the deployment of around 3,000 AI agents internally. Evans argued the company was moving toward a “100x organisation” in which employees primarily oversee and review the work completed by AI systems.

Despite growing enthusiasm around AI productivity, research findings remain mixed. A meta-analysis published in UC Berkeley’s California Management Review found no strong evidence linking AI adoption to major aggregate productivity gains. Other research from the National Bureau of Economic Research suggested that while AI can improve efficiency in some areas, perceived gains often exceed measurable real-world results.

Meanwhile, researchers at MIT studying AI agents concluded that many systems still struggle to produce human-quality work consistently. Their analysis suggested that current AI models may only achieve minimally sufficient performance on most text-based tasks sometime around 2029, with true human-level reliability likely requiring additional years of progress.

Other studies have pointed to another issue: management bottlenecks. Research published in the Harvard Business Review found that when AI tools allow workers to produce more content and ideas faster, decision-making pressure increasingly shifts upward to executives who must review, approve, and coordinate that work.

As companies continue to restructure around AI, some analysts believe the biggest challenge may not be whether AI can replace employees, but whether organisations themselves are prepared to handle the operational complexity and potential chaos that rapid automation could create.