AI Psychosis: Silicon Valley CEOs Are Firing Thousands Based on Tech That Isn't Ready
Tech layoffs in 2026 are on pace to surpass last year's total — and AI is the stated reason. More than 115,430 workers across 152 companies were cut in just the first five months of the year, already approaching the 124,636 total for all of 2025, per TechCrunch. Executives are replacing people with AI agents they claim can do the same work. The problem: independent researchers say current AI systems can't reliably handle most of those tasks.
The demo gap
Aaron Levie, CEO of cloud platform Box, has a name for what's driving these decisions: "AI psychosis." His diagnosis is that many executives are making sweeping workforce decisions based on polished demos rather than production reality. CEOs see AI generate a slick contract draft or produce a line of code and assume the hard, messy work underneath is equally solved. It isn't. They're typically the furthest removed from the day-to-day grind — the bug hunts, the hallucinated libraries, the hours spent correcting AI-generated legal documents full of invented clauses.
CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI.
— Aaron Levie (@levie) May 24, 2026
So when they play with AI, they see the happy path results, often not considering the next 10 or 20 things that have… https://t.co/ne5mvJ4Rgx
ClickUp is the clearest example of this thinking in action. CEO Zeb Evans announced a 22% staff cut after deploying roughly 3,000 internal AI agents — a ratio of three agents for every one employee, reports TheNextWeb. Evans framed the cuts as structural necessity, not cost reduction, and introduced $1 million salary bands for what he calls "AI-native" talent. The vision: humans manage algorithms, not do the work themselves.
What the research actually shows
That vision runs ahead of what AI can currently deliver. MIT researchers tested 41 AI models against 11,000 text-based workplace tasks drawn from Labor Department classifications. The models barely cleared minimum competency thresholds — and the MIT team projects AI will only reach 80–95% competency on those tasks by 2029, per Axios. A meta-analysis in the California Management Review found no consistent link between AI adoption and measurable productivity gains. A Gartner survey found 80% of companies using autonomous AI cut jobs — but those reductions aren't translating into meaningful financial returns.
There's a structural irony buried in all of this. When AI generates ten times the documents and decisions, someone still has to check them. Harvard Business Review notes that the approval chain becomes the new bottleneck: executives end up drowning in AI output rather than freed from work. The net result is less a productivity boom and more a flood of mediocre content that still needs a human eye.
The bottom line
The gap between what AI demos show and what deployed AI actually delivers is real and documented. Companies cutting headcount today on the assumption that AI will close that gap by tomorrow are making a bet that the research doesn't yet support. Workers paying the price for that bet deserve to know the evidence is thinner than the press releases suggest.