The tech sector is going feral. It echoes the cloud-computing wild west—think runaway costs, early-stage panic—but it also feels unprecedented. Record revenues coexist with mass firings. It is a weird dichotomy.
Aaron Levie, Box founder, called it what it is: AI psychosis. He isn’t being subtle. On X, he laid it out for 2.7 million people to read. CEOs are detached from reality. They play with a tool. Build a prototype. Generate a contract. Then they jump to the conclusion that agents will do all the actual work.
They never touch the messy stuff. They don’t see the code breaks. They don’t find the calls to hallucinated libraries. They aren’t the ones training models on specific contract quirks. They sure don’t spend days hunting for sneaky legal clauses. Levie says executives are too far from the last mile. They lack the granularity to understand automation limits.
But lack of understanding doesn’t stop the orders.
Levie isn’t an AI hater, for the record. He bets on it. He calls “headless software” the future. He backs AI startups heavily. His advice to peers is straightforward: Use AI. Use it “a ton.” Break your assumptions. Appreciate the upside. Also appreciate the real work still required.
Are CEOs listening? The data suggests not.
Just five months into 2026. Look at the numbers from Layoffs.fyi: 115,438 fired. Compare that to all of 2023 (which had 124,036 layoffs across a broader period). Tech companies are cutting heads fast. And when asked why? Almost everyone points to AI.
Many critics call it AI washing. A convenient label. They attribute past cuts to future efficiency. The real drivers are often financial metrics, not algorithmic genius.
Zeb Evans at ClickUp is the poster child for this disconnect. He cut 22% of his staff. Why? To deploy 3,000 internal AI agents. He insists this isn’t about cost-cutting. It’s about creating a “100x org.” People reviewing agent output, nothing more. He believes in this utopia.
The research disagrees. Hard.
A UC Berkeley meta-analysis in October found no robust relationship between AI adoption and productivity gains. Period.
The NBER study in March was kinder, but noted a productivity paradox : perceived gains look bigger than actual ones. The MIT researchers tested thousands of agents. Result? They don’t hit human-quality standards yet. At the current pace of LLM improvement, these tools might handle most text tasks with 80-95% success by 2029. And that’s just minimally sufficient. Base competence. Maybe three years away. Beating humans? That is further off.
There is a bottleneck, too. Harvard Business Review noted something sharp here: if everyone produces more with AI, the choke point moves upward. Executives must approve everything. What happens then? We saw hints in 2024. When everyone can act instantly, things get messy. Fast.
Are CEOs ready for a bottleneck at their feet? If the answer is no, psychosis leads nowhere good. Chaos is the only guaranteed output.
“The productivity paradox… in which perceived productivity gains are larger measured productivity gains.”
