A single tweet drew 252 likes. "I strongly believe there are entire companies right now suffering from AI psychosis." It was one line posted by Mitchell Hashimoto, co-founder of HashiCorp. No polished slide deck, no lengthy analysis. One short sentence resonated with hundreds of Silicon Valley developers and founders.
Hashimoto built open source projects that have earned over a million stars on GitHub and is known as the architect of Terraform and Vagrant. He led HashiCorp himself until its sale to IBM. The fact that he is not an AI skeptic or doomsayer but a core figure in Silicon Valley's technical infrastructure is what gives the remark its weight. This is testimony from someone who has watched AI adoption up close, from the inside.
What that remark means for the Korean market right now is worth unpacking on its own.
How AI Panic Swallows Decision-Making
Psychosis describes a state in which one's grip on reality has broken down. Borrowing a psychiatric term for corporate management may feel like a stretch, but Hashimoto's point is simple. He means decisions made under a tangle of fear that the company will fail without AI, an inflated sense of crisis that competitors are already pulling ahead with it, and uncritical faith that AI holds the right answers. The term applies when such decisions stop being one-offs and settle in as the operating mode of the entire organization.
The 81 comments on Hacker News filled up with firsthand accounts attached to that single line. Few named specific companies, but the patterns repeated. One developer wrote that their entire team had started committing AI output without reviewing it. A product manager described asking an AI for a strategic decision and reporting its answer to executives verbatim. A consultant cited a client who canceled a hire solely because an AI had recommended against it.
These accounts share one thing: AI has been elevated from assistive tool to final arbiter. Not a structure where humans set the direction and AI supports it, but one where AI sets the direction and humans stamp the approval. The difference looks small on the surface, but as it compounds, it changes the organization's very capacity to form judgments of its own.
The Korean market is no different. In the startup scene, being "AI-native" became the benchmark for competitiveness some time ago. Investors ask about AI strategy, and job postings list AI fluency as a baseline requirement. Companies bring in subscription tools to boost productivity, and executives invoke AI in every meeting. How many organizations have actually sorted out how much of this began from genuine need—and how much from the fear of falling behind?
Behind Rapid Adoption, the Judgment Circuit Quietly Atrophies
What makes AI psychosis frightening is that its symptoms look like performance. Fast adoption, heavy AI spending, company-wide prompt training, automated workflows, a newly minted AI task force. From the outside, these are hard to distinguish from the hallmarks of an innovative company. Inside, something entirely different may be unfolding.
No decision gets made without AI. People asked for their judgment delegate that judgment to the tool. When a result looks off, someone says, "Well, the AI said so." In strategy meetings, the output of a tool that knows none of the context becomes the presentation deck. The team sends signals that something needs a second look, but those signals get buried under the unspoken pressure to never slow down AI adoption.
As this deepens, the organization's judgment circuitry itself withers. Intuition born of human experience loses its footing, and the confident sentences of AI output fill the space. AI answers in the same tone even when it is wrong. Its output betrays no uncertainty. The moment AI output carries more authority in the conference room than a colleague raising an objection, that organization has already surrendered part of its judgment.
The opposing view deserves a fair hearing here. With the AI technology cycle moving this fast, maintaining an aggressive adoption pace is itself a survival strategy. The gap now felt by organizations that hesitated on AI two years ago is real. There is also a legitimate concern that the "psychosis" frame, if misread as a case against AI adoption itself, could invite even greater losses. Similar cautionary discourse circulated during the internet transition and the mobile transition, and many of the companies that took those warnings literally were ultimately pushed aside.
But Hashimoto is not taking aim at speed. He is taking aim at judgment. Adopting AI quickly is not the problem; the problem is a state in which no one is left inside the organization to think rationally about the direction and manner of that adoption. The more sophisticated the tool becomes, the more important the person who reads its output critically and integrates it into decisions. That role never leaves human hands, no matter how far the tool advances.
Find Out Who Is Actually Making the Decisions at Your Company
It may look like a big-company problem, but AI psychosis operates regardless of scale. If anything, it progresses faster in solo businesses and small teams. There is no external audit function, few colleagues to challenge a decision, and the pace of subscription spending easily outruns the pace of verifying what that spending delivers.
Here are a few self-diagnostic questions. In the past month, how many times did you look at an AI output, decide "this doesn't seem right," and not use it? If the answer is zero, you are closer to passing AI output through than to actually using it. A track record with no rejections may be a sign that judgment isn't functioning.
When you adopted the AI tools you use now, did you answer the question "what specifically becomes painful without this?"—or did you start from "it feels like falling behind not to use it"? Adoption born of fear never produces criteria for measuring effectiveness. Six months later, you still can't tell whether the tool is useful or useless.
If the time you spend reading primary sources in your own field has shrunk since adopting AI, the foundation you need to assess the validity of AI output is likely shrinking along with it. Writing good prompts and judging the soundness of what comes back are entirely different skills. The better you get at using AI, the more you need to grow the expertise that lets you keep its output in check. In many cases, though, the two move in opposite directions.
The real question in AI adoption is not which tools you use. It is what posture the person using them maintains. The belief that better tools, used faster, automatically produce better results may itself be the earliest symptom of AI psychosis.
What matters is whether someone in the organization notices when the AI is wrong. If that someone is you alone, you need to check periodically that this judgment is still intact. The more you depend on the tool, the more you need a routine for verifying that you are pointing it in the right direction.
AI psychosis is not a problem of people who use AI. It is a question of whether anyone in the organization remains willing to say "this doesn't seem right" when the AI is wrong. That capacity belongs to the realm of attitude, not technology. Better models will keep arriving, but the judgment to recognize when a tool is being used in the wrong direction will remain, to the very end, a human responsibility.
