Tech layoffs in 2026 are already approaching the total count from all of 2025 - and a meaningful share of those cuts are being justified not by revenue pressure or market slowdowns, but by AI agent adoption.
Box founder Aaron Levie has a name for the pattern: "AI psychosis." In a recent TechCrunch discussion, Levie argued that the executives deciding AI can replace entire departments are often the people least familiar with what those departments actually do day to day. ClickUp recently cut 22% of its workforce citing AI agents as the replacement plan - a significant bet from a project management company that has staked a large part of its product roadmap on agentic automation.
The layoff calculus sounds clean in a boardroom: deploy AI agents, reduce headcount, lower costs. The problem is that most knowledge work is messier than it looks from the outside. A customer success manager isn't just answering tickets - they're holding institutional memory about which clients are flight risks, reading subtext in emails, and escalating issues before they become churn. An AI agent can handle the tickets. The judgment layer is harder.
The Disconnect Between Vision and Reality
The "AI psychosis" framing points to something real in how automation decisions get made at tech companies. The people who control headcount decisions - founders, CFOs, board members - tend to interact with work outputs, not work processes. They see that something got done, not the invisible scaffolding of human coordination, context, and recovery-from-failure that made it happen.
This isn't a new problem. Automation waves in manufacturing and logistics ran into the same pattern: the work looked simpler than it was, early automation hit edge cases the system couldn't handle, and companies quietly rehired or restructured. The difference now is that language models are genuinely capable at a wide surface area of knowledge tasks, which makes overconfidence harder to diagnose before the damage is done.
The current wave of AI-motivated layoffs is happening faster than most companies can validate whether the replacement actually works. ClickUp made its 22% cut announcement this spring. The operational outcomes will show up in the next few quarters.
What This Means for Teams
For people doing jobs that look like prime AI automation candidates - content production, customer support, data analysis, coding - the honest picture is mixed. Some of that work genuinely will be compressed by AI tools. The portion that involves judgment, relationship context, and handling novel situations is more durable than the blanket "AI will replace it all" framing suggests.
The more immediate practical risk isn't obsolescence - it's working at a company that makes a poorly-calibrated automation bet, takes the headcount hit, and then has to rebuild capability six to twelve months later. That's a career disruption even if your individual skills remain valuable.
Levie's "AI psychosis" label is useful precisely because it names the organizational failure mode, not just the technology risk. The problem isn't that AI agents are too capable - it's that decision-makers are systematically overestimating how well-defined the work is that they're trying to automate.