An attorney at a legal tech startup quit her job this year. She did not cite a pay dispute, a difficult colleague, or a lack of advancement. She said, simply: “I quit 100 percent because of the AI use.”
Her boss had started using ChatGPT to write Slack messages and company emails. Then he started using it to make structural decisions about the organization. Then – and this is the part that made her leave – he started using it to decide who to hire and fire. When she pushed back on the AI’s conclusions, she became the problem.
She is not alone. A Reddit post titled “My Boss Has AI Psychosis and We’re F*****” went viral this month, collecting thousands of upvotes from workers who recognized the pattern immediately: a manager who runs every decision through a chatbot, shares the outputs as directives, and blames employees when reality diverges from what the AI projected. The post gave a name to something surveys now confirm is widespread – and that name is starting to stick.
The term got its sharpest definition from Aaron Levie, CEO of Box, who wrote in May that “CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen.” When executives play with AI tools, Levie argued in Fortune, they see the smooth demonstration, not the ten or twenty additional steps required to translate that output into actual results – and they blame the people doing those steps when the AI’s confident projection fails to materialize.
The viral post described exactly that dynamic. The poster’s boss had developed what they called an unhealthy dependency on Claude – Anthropic’s AI assistant – sharing its outputs as authoritative projections and, when employees fell short, concluding that the employees were the problem rather than the projections. “It’s not humanly possible,” the poster wrote, describing targets derived from AI calculations no team could actually meet. “I’m watching him get deeper and deeper into the hole. It’s not good for his mental health, it’s not good for any of us.”
The less dramatic but more widespread version plays out in inboxes and team channels daily. A non-profit employee described receiving emails from their manager within minutes of a meeting ending, full of “ideas” that were, as the employee put it, “clearly copy-paste from ChatGPT” and “often do not reflect the reality of how the program would work in the context of our org.” A project manager at a web design firm said their job had been “largely reduced to reviewing a never-ending slog through AI slop” – content that looks polished at first glance but falls apart on contact with the actual requirements of the work. Another worker described a supervisor who routinely fed conversation transcripts into ChatGPT to ask whether they had handled the interaction correctly, then acted on the chatbot’s verdict.
The anecdotes have numbers behind them. A 2026 survey of American workers found that 44 percent believe AI is doing more harm than good in their workplace, compared to 31 percent who see it as a net positive. Forty-seven percent said AI tools introduced at their company had increased their workload rather than reduced it – not because they were doing more valuable work, but because they were spending time correcting AI-generated errors, managing automated outputs that needed human review, and learning systems that created new tasks faster than they eliminated old ones. Futurism reported on the growing pattern of managers barging staff with nonsensical AI directives, perpetual pivoting, and structural decisions made without human judgment intervening.
The mandate often flows in one direction. AI-driven restructuring has reshaped major tech employers, with companies pointing to AI strategy as the rationale for thousands of job cuts. A Fortune survey found that CEOs are heavily pushing AI adoption onto their employees while barely using the tools themselves – creating an enforcement asymmetry in which workers are measured against AI-generated benchmarks set by executives who have not experienced the friction of actually operating those tools day to day. Federal policy in 2026 has reinforced the enterprise AI push, with few guardrails on how those mandates translate into individual workplace expectations.
Workers have started pushing back, sometimes covertly. A survey found that 44 percent of Gen Z employees admitted to actively sabotaging in-house AI deployments – introducing errors, slowing adoption, or undermining integrations – to protect their job functions. The sabotage partly reflects self-interest, but it also reflects a rational response to a system in which AI outputs function as targets that humans must meet, rather than tools that humans direct.
UC Berkeley researchers studying workplace AI adoption found a counterintuitive pattern: workers who embraced AI most enthusiastically showed the earliest signs of burnout. Tools that made more feel doable caused to-do lists to expand to fill every hour the tools freed up, with work bleeding into lunches and evenings. The productivity gains executives cited in board presentations were, for many employees, experienced as a faster treadmill rather than a shorter workday.
What no workplace policy or HR framework has addressed is the accountability question the attorney’s resignation raised. If a manager makes a firing decision based on a ChatGPT recommendation, and the fired employee can show the recommendation was factually wrong, the chain of responsibility is unclear. The chatbot’s developer has no employment relationship with either party. The manager exercised judgment, technically, by choosing to trust the tool. The employee is gone either way.
No one has written that policy yet. The chatbots are already in the room.
