Tech CEOs are apparently suffering from AI psychosis

There is simply a definite wildness successful the tech manufacture these days that some mimics erstwhile eras of ample changes, similar unreality computing (runaway costs successful the aboriginal days), and is similar thing we’ve ever seen earlier (record revenues accompanied by wide layoffs).

A mentation doing the rounds attempts to explicate the phenomenon: Tech executives, particularly CEOs, are collectively suffering from delusions of grandeur acknowledgment to AI. And astatine slightest 1 tech CEO has said truthful retired loud: Box laminitis Aaron Levie.

“CEOs are uniquely prone to AI psychosis due to the fact that they’re sufficiently distant from the past mile of enactment that inactive has to hap to make astir worth with AI,” Levie wrote connected X.

CEOs “play with AI,” make a prototype, oregon make a contract, to usage Levie’s examples, and past marque the leap to believing agents tin bash the work.

But these top-level executives aren’t the radical who person to reappraisal code, observe bugs, and place calls to hallucinated libraries earlier bundle is deployed. They aren’t liable for grooming AI models connected a company’s idiosyncratic declaration terms, nor bash they person to walk days combing done contracts to find sneaky terms, arsenic Levie indicates.

In different words, Levie’s mentation posits, CEOs don’t truly recognize processes good capable to cognize what truly tin and can’t beryllium automated. But that deficiency of cognition doesn’t halt them from acting connected their beliefs.

It’s important to enactment that Levie is not an AI hater. Quite the opposite. He mostly posts AI positivity connected X to his 2.7 cardinal followers, penning blogs titled, “Headless bundle is the future” connected however bundle built for AI agents is the mode forward. He besides puts his wealth wherever his rima is, backing AI startups arsenic an progressive angel investor.

So what are CEOs to bash instead? Levie advises CEOs to usage AI “a ton” to truly spot what it tin and can’t do, “and travel retired the different broadside with an appreciation for some the upside and the existent work.”

I person capable religion successful humanity to judge that determination are CEOs retired determination attempting to bash conscionable that, but close now, they look to beryllium successful the minority.

In lone the archetypal 5 months of 2026, the tech manufacture has already had astir arsenic galore layoffs arsenic successful each of 2025: 115,430 radical person been fired from 152 tech companies truthful acold successful 2026, compared to 124,636 radical fto spell by 275 companies successful 2025, according to manufacture layoff tracker Layoffs.fyi.

And the bulk of companies person pointed to AI arsenic a crushed for cutting these jobs. Many reason that the biggest tech companies are AI washing, oregon crediting AI productivity gains successful the past oregon future, erstwhile different concern decisions and metrics are truly driving the cuts.

Still, immoderate of these stories are surprising. Zeb Evans, the CEO of task absorption and productivity bundle startup ClickUp, proudly declared connected X that helium had laid disconnected astir a 4th of his employees — 22% — aft rolling retired astir 3,000 AI agents to bash interior work.

Evans swore this wasn’t done to trim costs. Instead, helium wants a workforce composed of radical who tally AI agents and walk their days rapidly reviewing the agents’ work. He believes this volition make a “100x org,” arsenic helium calls it.

While AI tin beryllium a precise utile tool, the information connected AI and productivity doesn’t enactment specified assumptions. By miles.

A meta investigation of different probe published successful October successful UC Berkeley’s California Management Review recovered “no robust narration betwixt AI adoption and aggregate productivity gain.”

Research published successful March by the National Bureau of Economic Research did conclude that AI adoption improved productivity, but noted “a productivity paradox, successful which perceived productivity gains are larger than measured productivity gains.”

After creating thousands of agents to enactment connected tasks, researchers astatine MIT concluded that agents conscionable aren’t doing human-quality enactment yet successful galore cases. They foretell astatine the existent complaint of LLM improvement, models volition “be capable to implicit astir text-related tasks with occurrence rates of, connected average, 80%–95% by 2029 astatine a minimally capable prime level.”

In different words, AI is connected way to execute astatine basal competence connected astir tasks successful astir 3 years. These researchers judge agents volition request different fewer years to outperform humans.

Meanwhile, probe published successful the Harvard Business Review showed that erstwhile everyone is utilizing AI to nutrient much stuff, the bottleneck simply shifts to executives. Their enactment awaits the radical that indispensable authorize each the worldly everyone is producing. If everyone is empowered to act, past from what OpenAI experienced past year, we tin archer that things whitethorn get retired of control.

Are CEOs acceptable for that? If not, the astir definite result of the ongoing CEO AI psychosis volition simply beryllium organizational chaos.