A peculiar syndrome is sweeping through the upper echelons of the tech industry. It has no official medical name, but Box CEO Aaron Levie has given it a fitting label: AI psychosis. In a series of posts on X, Levie described how top executives are increasingly convinced that artificial intelligence can perform complex work with little human oversight, even as evidence mounts that the technology remains far from reliable.

The condition manifests in a specific pattern: CEOs use AI to generate prototypes, draft contracts, or create code demos. Impressed by the initial results, they leap to the conclusion that entire workflows can be automated end-to-end. But as Levie pointed out, these executives are insulated from the messy details of execution. They rarely review code for hallucinated libraries, debug unexpected outputs, or train models on idiosyncratic company data. The last mile of work—the grunt labor that transforms a promising demo into a production-ready system—remains invisible to them.

The layoff wave of 2026

The consequences of this disconnect are stark. According to Layoffs.fyi, a tracker of job cuts in the technology sector, 115,430 people have been laid off from 152 tech companies in the first five months of 2026 alone. That figure nearly matches the 124,636 layoffs recorded across 275 companies during all of 2025. The pace is accelerating, and a growing number of employers are explicitly citing AI as the reason for reducing headcount.

One of the most striking examples came from Zeb Evans, CEO of ClickUp, a project management software startup. In a public declaration on X, Evans announced that he had laid off 22 percent of his workforce after deploying approximately 3,000 AI agents to handle internal tasks. He insisted the move was not about cost-cutting, but about building what he called a '100x org'—a company where humans primarily supervise AI agents. Critics quickly noted that such a drastic reduction in staff would inevitably shift oversight burdens back onto remaining employees, creating a bottleneck at the executive level.

What the research actually says

Despite CEO enthusiasm, rigorous academic studies paint a far more cautious picture. A meta-analysis published in October 2025 in UC Berkeley's California Management Review examined multiple studies on AI and productivity. It found 'no robust relationship between AI adoption and aggregate productivity gain.' In other words, companies that invested heavily in AI did not outperform those that did not—at least not in measurable economic terms.

The National Bureau of Economic Research offered a slightly more optimistic view in a March 2026 paper. Its authors concluded that AI adoption did improve productivity, but they identified a 'productivity paradox'—managers perceived larger gains than actually materialized. This gap between belief and reality is precisely the cognitive bias Levie warns against.

Perhaps the most detailed forecast comes from researchers at the Massachusetts Institute of Technology. They created thousands of AI agents to perform text-related tasks and measured their performance. The conclusion: current large language models are not yet capable of producing human-quality work in many cases. The researchers predict that by 2029, models will achieve 'minimally sufficient quality' on 80 to 95 percent of tasks, but truly outperforming humans will take several more years after that.

The bottleneck moves upstairs

A study published in the Harvard Business Review highlighted an unintended consequence of widespread AI adoption. When every employee uses AI to generate more content, proposals, and analysis, the approval chain becomes the choke point. Executives who are already overwhelmed find themselves drowning in output that requires human judgment. The authors noted that OpenAI itself experienced chaos in 2025 when a surge of AI-generated work bypassed traditional safeguards, leading to embarrassing public failures.

Levie's diagnosis implies that CEOs are not malicious—they are simply uninformed about the actual mechanics of the tools they champion. He recommends that executives use AI 'a ton' themselves, but with a skeptical eye. 'Come out the other side with an appreciation for both the upside and the real work,' he wrote. Yet his warning may go unheeded in a culture that rewards bold predictions over careful implementation.

Historical parallels

The current AI mania echoes previous technological infatuations. During the early days of cloud computing, companies often racked up runaway costs because they assumed that migrating to the cloud would simplify everything. They discovered that managing cloud infrastructure required new skills, monitoring, and governance. Similarly, the dot-com bubble of the late 1990s was fueled by executives who believed that internet technology would magically transform business models overnight. Many of those companies collapsed when the complexity of execution caught up with them.

AI psychosis may be the most dangerous iteration yet because the technology is so opaque. When a CFO signs off on layoffs based on an AI optimism that has no basis in data, the damage extends beyond individual careers. It erodes trust in management and creates organizational chaos. As Levie puts it, the most likely outcome of this executive delusion is not a leaner, more efficient tech industry, but a series of avoidable breakdowns.

Source: TechCrunch News