Foreign media commentary suggests that the recent AI boom in the tech industry is pushing some corporate executives toward overly optimistic assessments. The article cites Aaron Levie, founder of Box, who noted that many CEOs, being distant from frontline operations, tend to equate AI demonstrations directly with the ability to broadly replace real-world workflows.
Executives see the demo, not the implementation.
Levie stated on social media that CEOs often personally experience AI by creating prototypes, generating contracts, or running simple workflows, then conclude that agents are ready to take over substantial work. However, those truly responsible for deployment must still review code, fix bugs, identify errors caused by model hallucinations, and handle the complex details of internal corporate contracts, processes, and data.
The article states that such misjudgments do not stem from opposition to AI. On the contrary, Levie has long been an enthusiastic supporter of AI and has invested in AI startups. His core argument is that the issue is not that AI lacks value, but that management often underestimates the human effort and time required to turn tools into reliable productivity.
The pace of layoffs has approached the total for all of last year.
The article cites data from Layoffs.fyi, stating that in the first five months of 2026, 152 tech companies laid off 115,430 employees, nearing the full-year 2025 total of 124,636 layoffs across 275 companies. The report notes that many companies cite AI as one reason for layoffs, but the underlying drivers may extend beyond technological advancement.
ClickUp CEO Zeb Evans publicly stated that after deploying approximately 3,000 AI agents to handle internal workflows, the company reduced its workforce by about 22%. He described this move not merely as a cost-cutting measure, but as an effort to restructure the team to focus on managing AI agents and rapidly reviewing their outputs.
The findings do not support radical alternatives.
However, the article notes that multiple studies have not reached similarly aggressive conclusions. A review study released in October last year by the University of California, Berkeley found no robust relationship between AI adoption and overall productivity gains. A study by the National Bureau of Economic Research in March this year concluded that while AI does enhance efficiency, subjective perceptions often exceed actual measured results.
Research from MIT on agents performing tasks also shows that, in many scenarios, agents still cannot consistently achieve human-level quality. Researchers anticipate that, at the current pace of large model advancements, models may be able to complete most text-related tasks at a “minimum acceptable quality” by 2029, but it will take considerably longer to reliably surpass human performance across a broader range of tasks.
The article concludes that if management continues to restructure the organization based on demo performances rather than actual implementation capabilities, the result may not be a leap in efficiency, but rather an accumulation of approvals, chaotic execution, and organizational imbalance.