On May 22, 2026, ClickUp CEO Zeb Evans posted on X that he’d cut 22% of his workforce. Not because the company was bleeding money. Not because the board forced his hand. Evans said he’d deployed roughly 3,000 AI agents across internal operations and concluded that the humans doing that work were no longer necessary. It was a textbook case of what’s now being called AI psychosis.
Evans called the restructured company a “100x org” and offered the remaining employees salary bands reaching $1 million per year to stay and manage the bots. The reaction was swift. Tech Twitter split between admiration and horror. Investors debated whether this was vision or recklessness. And then Box CEO Aaron Levie posted the diagnosis that stuck: CEOs, he wrote, are suffering from “AI psychosis.”
AI psychosis is a term coined by Box CEO Aaron Levie in May 2026 to describe the overconfident, distorted belief among executives that AI can automate roles and workflows they don’t personally understand, leading to sweeping decisions like mass layoffs based on incomplete evidence of what AI can actually do.
The term caught fire because it named something everyone in tech had been watching but nobody had labeled. In five months, 142,000 tech workers had lost their jobs. The bulk of companies cited AI as the reason. And a growing body of research suggested the productivity gains justifying those cuts might not exist yet.
Last updated: June 2026
Quick answers
Who coined AI psychosis?
Box CEO Aaron Levie coined the business concept of “AI psychosis” on May 24, 2026, in a post on X. Levie described CEOs as “uniquely prone to AI psychosis” because they’re “sufficiently distant from the last mile of work” to see AI’s limitations firsthand. The term is separate from the clinical concept of chatbot-induced psychosis, first described by Danish psychiatrist Soren Dinesen Ostergaard in 2023.
Is AI psychosis a real medical condition?
There are two uses of the term. The business meaning, coined by Aaron Levie, describes CEO overconfidence in AI. It’s not a medical diagnosis. Separately, the clinical term “AI psychosis” or “chatbot psychosis” describes delusional experiences triggered by heavy chatbot use. Wikipedia now has a page on chatbot psychosis as a mental health phenomenon, but it’s not yet recognized in the DSM.
Who coined AI psychosis?
Aaron Levie, the CEO and co-founder of Box, posted the term on X on May 24, 2026. The full post: “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. So when they play with AI, they see the happy path results, often not considering the next 10 or 20 things that have to happen to get sustainable results from agents.”
The post went viral. TechCrunch, Fortune, Fast Company, and Inc. all published pieces within days. TechCrunch’s Julie Bort noted that Levie “said the quiet part out loud” about how tech leaders were making sweeping workforce decisions based on limited hands-on experience with AI tools.
What made Levie’s critique credible is that he’s not an AI skeptic. He runs a $1.18 billion revenue company (Box’s fiscal 2026 numbers, per SiliconANGLE) that’s betting heavily on AI. Box reported Q1 2026 revenue of $305.9 million, up 11% year over year, with AI-powered data extraction becoming what Levie calls a “killer app” for enterprise customers. He’s also an active AI angel investor. His critique came from inside the house.
Levie’s specific insight was about the “last mile.” CEOs prototype something with an AI tool, see it produce a draft contract or a working code snippet, and extrapolate that the tool can handle the full job. But they’re not the ones who review code for hallucinated libraries, train models on a company’s specific contract terms, or spend days catching errors in AI-generated output. The gap between the demo and the deployment is where AI psychosis lives.
Why are CEOs prone to AI psychosis?
The answer is structural, not personal. CEOs sit at the top of information hierarchies where two forces combine to distort their perception of AI’s capabilities.
The first is distance from execution. A CEO who asks ChatGPT to draft a memo and gets a coherent response in 10 seconds experiences a compressed, curated version of what AI can do. They don’t see the prompt engineering required to get consistent results, the fact-checking needed to catch fabricated citations, or the domain expertise necessary to evaluate whether the output is actually correct. A 2025 Rev survey of more than 1,000 AI users found that heavy users encounter 3x more hallucinations and spend nearly 10x longer getting satisfactory answers than casual users. The more you use AI, the more you see its cracks. CEOs often don’t use it enough to reach that stage.
The second is incentive alignment. Wall Street has rewarded AI narratives aggressively in 2026. When Meta announced 8,000 job cuts to fund its $135 billion AI capex plan, the framing wasn’t “we’re firing people.” It was “we’re restructuring for AI.” Salesforce CEO Marc Benioff cut 4,000 customer support roles with similar messaging. Deutsche Bank analysts wrote in January that “AI redundancy washing will be a significant feature of 2026.” OpenAI CEO Sam Altman acknowledged both dynamics exist, noting “some AI washing where people are blaming AI for layoffs they would otherwise do.”
The third factor is competitive panic. When one CEO announces an AI-driven restructuring, board members at competing companies start asking why their company hasn’t done the same. ClickUp’s Zeb Evans didn’t just cut jobs; he published the “100x org” thesis as a public playbook. That puts pressure on every other productivity software CEO to respond with their own AI strategy. Fear of being left behind accelerates decisions that haven’t been pressure-tested.
How many tech workers have been laid off because of AI in 2026?
The headline number is staggering: 142,000 tech workers lost jobs across more than 150 companies in the first five months of 2026, according to TechTimes. That already approaches the 2025 full-year total of 124,636 layoffs across 275 companies, per Layoffs.fyi.
The biggest cuts by company: Oracle eliminated up to 30,000 positions (roughly 20% of its workforce). Meta cut 8,000 (10% of staff). Amazon reduced 16,000 corporate roles in Q1 alone. Salesforce dropped 4,000 customer support jobs. ClickUp cut 290 people, or 22% of its 1,300-person team.
Nearly 48% of tech layoffs through April 2026 were explicitly attributed to AI and workflow automation, according to Tom’s Hardware. But that attribution deserves scrutiny. Oxford Economics concluded in January 2026 that firms “don’t appear to be replacing workers with AI on a significant scale,” suggesting some companies dress up headcount reductions as AI-driven transformation to satisfy investors.
The data tells a split story. Stanford HAI’s 2026 AI Index found that employment for software developers aged 22 to 25 fell nearly 20% since 2024. These are entry-level workers doing the tasks AI handles best: boilerplate code, scripted tests, implementing well-specified features. But developers aged 30 and older at the same companies saw employment grow 6 to 12% over the same period. AI isn’t eliminating software jobs. It’s eliminating junior software jobs while making senior roles more valuable.
The distinction matters for founders. If you’re hiring, the junior talent market has shifted. If you’re building, the question isn’t whether AI replaces people but which layers of work it can reliably handle without supervision.
What does the research say about AI productivity gains?
The research mostly contradicts the narrative driving AI psychosis. Here are four findings every founder running an AI-first strategy should know.
A UC Berkeley meta-analysis published in the California Management Review in October 2025 found “no robust relationship between AI adoption and aggregate productivity gain.” That’s not “the gains are small.” It’s “we can’t reliably detect them in the data.” Individual case studies show wins, but the macro picture hasn’t moved.
An NBER study published in March 2026 did find productivity improvements from AI adoption but flagged “a productivity paradox, in which perceived productivity gains are larger than measured productivity gains.” Translation: companies think AI is helping more than the numbers show. That perception gap is the substrate AI psychosis grows in.
MIT researchers evaluated AI performance across more than 3,000 workplace tasks and projected that at the current rate, AI models will handle 80-95% of text-based tasks by 2029 at a “minimally sufficient” quality level. “Minimally sufficient” means the output is usable without edits. As of late 2025, models scored that bar on roughly 65% of tasks. The gap between 65% and 95% is a three-year gap. Companies firing people today are betting on 2029 capabilities arriving in 2026.
Harvard Business Review research published in May 2026 identified a bottleneck problem: when everyone uses AI to produce more output, the approval and review workload shifts upstream to managers and executives. The work doesn’t disappear. It moves.
Taken together, the research paints a consistent picture. AI produces real gains in specific, well-scoped tasks. Writing first drafts, generating boilerplate code, summarizing documents. But the leap from “AI helps with tasks” to “AI replaces people” requires ignoring the connective tissue that holds organizations together: context switching, institutional knowledge, cross-team coordination, judgment calls that don’t fit neatly into a prompt. Companies that treat AI as a tool for augmenting human judgment tend to see the best returns. Companies that treat it as a substitute for human judgment tend to produce the headlines about AI psychosis.
5 signs your company has AI psychosis
Levie’s framing is about CEOs, but AI psychosis can infect an entire organization. Here’s how to spot it in your own company, whether you’re the founder or employee number twelve.
1. You’re automating roles you haven’t personally done in years. If the CEO is cutting customer support because “AI can handle it” but hasn’t worked a support ticket since 2019, that’s AI psychosis. The people closest to the work usually have a more accurate read on what AI handles well and what falls apart at scale. Ask the team leads, not the board deck.
2. Your AI metrics measure output, not outcomes. “We generated 4x more content with AI” sounds impressive until you check whether anyone read it, whether it converted, whether it was accurate. Counting AI-produced documents without measuring their quality and impact is the corporate equivalent of confusing activity with progress. A shift toward agentic engineering means measuring what agents actually accomplish, not what they produce.
3. Your competitor made an AI announcement and now the board wants a response. Reactive AI strategy is a hallmark of the condition. ClickUp’s “100x org” announcement created pressure across the productivity software sector. But responding to a competitor’s narrative with your own layoffs is the corporate version of keeping up with the Joneses. Strategy based on what another company is doing with AI, rather than what your customers need, usually ends badly.
4. You haven’t budgeted for the “last mile.” Levie’s core insight: there’s a gulf between what AI produces and what’s ready for a customer. If your budget includes the AI tool but not the human review, error correction, edge-case handling, and model training required to make it work, you’ve priced for the demo, not the deployment. The Rev survey found that users with complex, multi-source prompts are 32x more likely to revise for hallucinations “very often” compared to simple prompts.
5. You’re describing layoffs as “transformation.” Language matters. When people lose their jobs, calling it an “AI-driven transformation” doesn’t change the human cost. Companies with genuine AI transformations typically redeploy workers into new roles, not out the door. If the only “transformation” is fewer people on payroll, that’s a layoff with better branding.
What should founders do instead?
Levie’s own advice is simple: use AI “a ton” until you develop an appreciation for both the upside and the real work involved. That’s a starting point. Here’s a more specific playbook, informed by what the research and the layoff data actually suggest.
Run pilots before restructuring. Before announcing that AI can replace a function, run a 90-day pilot where AI handles real tasks in that function while humans remain in place. Measure not just speed but accuracy, customer satisfaction, and edge-case failure rates. If the pilot doesn’t run 90 days, you don’t have enough data.
Budget for the quality gap. MIT’s research says AI currently passes the “minimally sufficient” bar on 65% of text tasks. That means 35% of the time, a human still needs to step in. Build that into your cost model. An AI co-founder handles some work, but the founder still does the thinking.
Hire for supervision, not just execution. The Stanford data shows that senior developers are more valuable in an AI-augmented workplace, not less. The same pattern will hold across functions. People who can evaluate, correct, and improve AI output are worth more than people who simply produce it. If your hiring plan only accounts for AI operators and not AI supervisors, flip the ratio.
Watch the bottleneck shift. The HBR research on managerial bottlenecks is a warning sign for lean startups. If three employees are now producing the output of ten thanks to AI, someone still has to review, approve, and integrate that output. Make sure your org structure accounts for where the work goes after AI produces it, or you’ll hit a different kind of gridlock.
Talk to your team before your board. The people closest to the work know where AI actually helps. Customer support agents know which tickets AI handles cleanly and which it botches. Engineers know which code suggestions they accept and which they rewrite entirely. The information that prevents AI psychosis lives at the operational level, not the executive level. If you’re building your AI-first company, ground your decisions in what the tools can actually do today.
Levie isn’t saying AI is overhyped. He’s saying the timeline is compressed in executives’ heads. The technology works. It’s getting better fast. MIT projects real competence by 2029. The psychosis isn’t believing AI will change work. It’s acting like it already has.
For founders building right now, the opportunity is real. AI tools let small teams punch above their weight in ways that would’ve been impossible two years ago. The founders who’ll win aren’t the ones ignoring AI or the ones firing everyone to replace them with agents. They’re the ones who understand exactly where the technology’s edge is today and build their operations around that reality, not the press release.
