[Digital Today reporter Jinju Hong] Criticism has emerged in the tech industry that chief executives are overestimating the real scope of automation possible with artificial intelligence (AI).

TechCrunch reported on May 27 that Box founder Aaron Levie (에런 레비) described recent tech industry trends by saying, “It looks like CEOs have collectively fallen into an AI delusion.”

In a post on social media platform X, formerly Twitter, Levie explained that executives are often far from day-to-day operations, making it easy to overestimate AI’s automation potential. “CEOs are especially vulnerable to delusion from AI,” he said. He added that many believe AI agents can replace real work after seeing only capabilities such as building prototypes or generating contract drafts.

He stressed that many steps still require human involvement even after AI generates an output. Tasks such as reviewing code before software deployment, fixing bugs and checking calls to non-existent libraries ultimately need skilled developers, he said. He also viewed training models to reflect company-specific contract terms or reviewing hidden clauses in contracts as far from full automation.

The remarks also intersect with the tech industry’s recent wave of large-scale layoffs. Layoffs.fyi said 152 tech companies cut a total of 115,430 jobs in the first 5 months of 2026. That is close to the 124,636 jobs cut by 275 companies over all of 2025. Critics say many companies cite AI to justify layoffs, but in reality often rebrand business slowdowns or cost-cutting as AI-driven productivity innovation.

The case of work management platform ClickUp illustrates the debate. CEO Zeb Evans (젭 에번스) disclosed that after deploying about 3,000 AI agents for internal work, it cut about 22 percent of its workforce. He described it not as simple cost-cutting but as “a transition to an organisation that manages AI agents and rapidly reviews results,” and presented a so-called “100x organisation” vision.

But research findings do not support claims that adopting AI leads directly to productivity innovation. California Management Review, which published a UC Berkeley meta-analysis, concluded there is “no robust relationship” between AI adoption and overall productivity gains. A National Bureau of Economic Research study also said AI can bring some productivity improvement but pointed to a “productivity paradox,” in which perceived expectations are larger than measurable results.

Projections about AI replacing on-the-ground work also remain limited. MIT researchers said that experiments with thousands of AI agents showed they failed to reach human-level quality in many tasks. Assuming the current pace of development in large language models, the researchers expected that around 2029 AI could perform most text-based work at 80 to 95 percent of a minimum practical quality standard. They also said it would take longer to reach a level superior to humans.

Some also warn that the spread of AI could create new bottlenecks. A study published in Harvard Business Review said that if all members of an organisation use AI to produce more outputs, executives with final approval authority could become a new bottleneck. That would mean organisational burdens could rise as approval backlogs surge.

Levie stressed that executives should not stop at a superficial experience of AI. “CEOs need to use AI a lot,” he said, adding that the process would help them understand not only AI’s potential but also the scope of human labour that is still required. As tech companies increasingly cite AI as a core rationale for layoffs and restructuring, warnings are growing that overconfidence in AI could instead lead to new organisational confusion.