On the day ClickUp announced it was letting go of 22% of its entire workforce, the official statement was spare: "AI agents will take over those roles." But Box co-founder Aaron Levie put a name to the trend — "AI psychosis." It describes a state in which an organization becomes so enamored of AI's possibilities that it loses track of how work actually gets done. What makes the condition more dangerous is that the people making the call are usually the ones who have never done the job themselves.

In 2026, the Reason for Layoffs Has Changed

As of the first half of 2026, the scale of tech-industry layoffs has already neared the full-year total for 2025. It isn't just that the pace has accelerated. The reasons have changed. In earlier rounds of cuts, the main justifications were an economic slowdown, overhiring, and cost efficiency. The 2026 layoff memos feature different words: "AI agents," "automation transition," "role redefinition."

ClickUp's announcement is a textbook case. For a project-management SaaS company, 22% is no small figure relative to its size. The company's logic is that AI agents can now handle the repetitive work its existing staff used to do. That premise isn't entirely wrong. Routine ticket handling, log triage, and internal document cleanup are tasks AI can absorb in large part.

The problem is where that judgment was made. Identifying which work is genuinely routine — and which only looks routine but isn't — requires asking the people who have actually done it. And the people in a position to decide on layoffs have, more often than not, never done the work themselves.

What AI Is Said to Do, and What It Actually Does

Here is the structure of the "AI psychosis" Levie described. You watch a demo of AI doing something, and you come to believe it will work exactly the same way in a real work environment. A demo is built from controlled inputs, ideal conditions, and predictable outputs. Real work is none of those things.

Consider the day of a support agent fielding customer inquiries. If 60% of the questions can be neatly categorized, the remaining 40% involve emotions that resist categorization, context-dependent judgment, and situations that demand tacit institutional knowledge. AI handles the 60%. But the decision-maker's announcement reads simply as "AI handles customer inquiries." The 40% hasn't disappeared — the people have. Who will handle that 40% isn't in the plan. Either the existing team absorbs it, or the customer experience quietly degrades, or a new job posting goes up a few months later.

There are counterarguments, of course. It isn't as though no AI transition has genuinely strengthened an organization's capabilities. Some companies have handed repetitive work to AI and freed the remaining staff to focus on higher-value work. Salesforce and some enterprise firms have reported cutting total labor costs after adopting AI while maintaining customer satisfaction. From this vantage point, AI-driven layoffs don't necessarily lead to a loss of organizational capability. The claim that organizations that pivoted quickly became more agile is also hard to dismiss.

But those success stories share a common condition. People who understood the frontline work were deeply involved in designing the transition, and the scope handed to AI was expanded gradually through experimentation. It wasn't a fast decision, a big announcement, and a sweeping layoff. Before designing a solution, they first looked at the problem through the eyes of the people who live with it every day. Whether you're planning a product or transforming an organization, reverse that order and the result changes.

Why This Pattern Keeps Repeating in Korean Organizations

When the conversation about adopting AI begins at a Korean company, it's rare for the frontline staff to be the ones who raise it first. Direction gets set from the top, a vendor demo goes up, and a cost-savings report makes its way to the executives. Along the way, the voice of the person who knows the actual work best grows fainter — because tacit knowledge that can't be expressed in numbers never makes it into the report.

This structure persists not out of bad intentions. The decision-makers see AI's promise and sincerely believe it's good for the organization. The frontline staff know how the transition will actually play out, but often lack the language to convey that persuasively upward. The demo's logic is clean; the exceptions on the ground are messy. The report picks the clean side.

If you're a solo operator or a director leading a small team, you'll more often have to make the AI-adoption call yourself. In that seat, the way to avoid "AI psychosis" is to cut the number of steps. First break down one team member's day in detail, run a two-week experiment on which parts can actually be automated, and only then draw conclusions. The decision to roll it out across the whole organization comes after that.

If you're a middle manager, what's needed most when AI adoption comes up is to lay out the frontline staff's actual work in concrete terms and carry it upward. A sentence like "AI can handle 70% of this work, but the remaining 30% still needs a person, for these reasons" has to make it into the report. Without that sentence, only the numbers remain.

There are three things to check. First, confirm whether the person making the AI-adoption decision has ever done the work in question firsthand. Second, if a sweeping, all-at-once transition is planned with no pilot period, be suspicious of the premise. Third, check whether there's a plan for who handles the remaining work, and how, after a layoff announcement. If there are no clear answers to these three, the decision may well have mistaken a demo for reality.

In 2026, the moment to decide whether to adopt AI has already passed. The question now is how to adopt it — and where to slow down. What ClickUp will only confront after letting 22% of its people go, organizations facing the decision now can see in advance. How well the person making that call understands the frontline work is what will divide the organization's next year.