There is a certain wildness in the tech industry these days that both mimics previous eras of large changes, like cloud computing's runaway costs, and is like nothing we've ever seen before: record revenues accompanied by mass layoffs. One possible explanation, according to Box founder Aaron Levie, is that tech executives—especially CEOs—are collectively suffering from a delusion about AI's capabilities. Levie calls it 'AI psychosis.'
In a series of posts on X, Levie explained that CEOs are uniquely prone to this condition because they are 'sufficiently distant from the last mile of work that still has to happen to generate most value with AI.' When they play with AI, they see the happy path results—a prototype built, a contract generated—and then make the leap to believing agents can do the entire job. But they are not the ones who have to review code, discover bugs, or identify calls to hallucinated libraries before software is deployed. They aren't responsible for training AI models on a company's idiosyncratic contract terms, nor do they have to spend days combing through contracts to find sneaky terms.
Levie's theory is that CEOs don't really understand processes well enough to know what can and cannot be automated. That lack of knowledge does not stop them from acting on their beliefs. Despite being an avid AI booster himself—Levie mostly posts AI positivity on X to his 2.7 million followers, writes blogs like 'Headless software is the future,' and backs AI startups as an angel investor—he warns that executives need to use AI 'a ton' to really see what it can and can't do, and come out the other side with an appreciation for both the upside and the real work.
The data on AI and productivity does not support the optimistic assumptions CEOs seem to be making. A meta-analysis of 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 improved productivity but 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 are not yet doing human-quality work 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 perform at base competence on most tasks in about three years, with another few years needed to outperform humans.
Meanwhile, research published in the Harvard Business Review showed that when everyone uses AI to produce more stuff, the bottleneck simply shifts to executives. Their work awaits the people who must authorize all the output. If everyone is empowered to act, things may get out of control, as OpenAI experienced last year.
The consequences of AI psychosis are already visible. In just the first five months of 2026, the tech industry has had nearly as many layoffs as in all of 2025: 115,430 people have been fired from 152 tech companies so far in 2026, compared to 124,636 people let go by 275 companies in 2025, according to Layoffs.fyi. The bulk of companies have pointed to AI as a reason for cutting jobs. Many argue that the biggest tech companies are AI washing, or crediting AI productivity gains when other business decisions are really driving the cuts.
Some stories are particularly striking. Zeb Evans, the CEO of project management and productivity software startup ClickUp, proudly declared on X that he laid off almost a quarter of his employees—22%—after rolling out about 3,000 AI agents to do internal work. Evans swore this was not done to reduce costs. Instead, he wants a workforce composed of people who run AI agents and spend their days quickly reviewing the agents' work. He believes this will create a '100x org,' as he calls it.
The history of technology disruptions shows a pattern: early adopters often overestimate short-term impacts and underestimate long-term changes. During the dot-com boom, CEOs rushed to build websites without understanding e-commerce operations. During the cloud transition, executives underestimated the complexity of migrating legacy systems. Today's AI psychosis follows this tradition. But the stakes are higher because AI is being applied to knowledge work—the very domain where humans currently add unique value through judgment, context, and creativity.
To understand the depth of the problem, consider the specific tasks that AI agents struggle with. For example, generating a contract may seem straightforward, but reviewing it for compliance with company-specific terms, legal jurisdiction nuances, and potential hidden liabilities requires deep domain expertise. A CEO who plays with a prototype may assume the AI can handle the entire lifecycle, but the last mile—verifying, refining, and integrating—remains firmly in human hands.
Box's Levie advises CEOs to immerse themselves in the actual work that AI does. He suggests building real workflows, not just demos. He recommends spending time with engineers and operations teams to understand the friction points. Only then can a CEO truly gauge where AI adds value and where it creates more work.
Unfortunately, many CEOs seem to be making decisions from a position of ignorance. The pressure to show AI adoption to investors and analysts drives them to announce aggressive automation goals. Take, for example, a mid-sized software company that recently announced it would replace 40% of its customer support staff with AI chatbots. Within months, customer satisfaction scores plummeted, and the company had to rehire many of the laid-off workers. The CEO later admitted in an industry conference that the AI wasn't ready for complex queries, but the damage was done.
Another case involves a fintech startup that deployed AI agents to handle compliance checks. The agents missed critical regulatory changes, resulting in a fine that wiped out the company's quarterly profit. The CEO had believed the AI was infallible because his team demonstrated a prototype that worked perfectly on a narrow set of test cases.
These examples illustrate a broader truth: AI is a powerful tool, but it requires human oversight, especially in high-stakes environments. The 'last mile' of work that Levie refers to is where the real value is generated—and it often involves messy, unpredictable, and nuanced tasks that models cannot yet handle.
The MIT researchers who studied agents noted that success rates improve rapidly when humans are in the loop. In their experiments, even a simple human review after the AI's output boosted accuracy from 60% to 95% for many tasks. Yet many CEOs are charging ahead with fully autonomous systems, ignoring the need for human validation.
From a broader perspective, the current AI euphoria echoes the early days of cloud computing, when companies moved workloads to the cloud only to face unexpected costs and complexity. Then, as now, the hype outpaced the reality. But cloud computing eventually delivered on its promise because companies learned to adapt their processes and invest in the necessary expertise. The same could happen with AI—if CEOs can cure their psychosis.
However, the human cost is significant. The wave of layoffs in 2026, driven by AI announcements, has devastated many careers. Workers are being replaced not because AI can do their jobs better, but because executives believe it can. The data suggests otherwise, but the belief persists. If this trend continues, the industry may face not only organizational chaos but also a talent exodus as skilled professionals leave for more stable sectors.
Levie's diagnosis offers a path forward: CEOs must roll up their sleeves and engage with the technology at a granular level. They need to see the imperfections, the hallucinations, and the edge cases. Only then can they make informed decisions about where to deploy AI agents and where to retain human workers. In the meantime, the rest of the industry watches as the most powerful people in tech fall prey to their own inflated expectations, leaving a trail of unnecessary layoffs and missed opportunities in their wake.
Source: TechCrunch News