When the Whole Company Goes Mad: Mitchell Hashimoto's Warning About 'Corporate AI Psychosis'

There’s a phrase quietly spreading across tech Twitter and Hacker News right now: AI psychosis. It started as shorthand for what happens to individuals who spend too much time talking to chatbots — slowly losing touch with reality as the model agrees with everything they say. Then HashiCorp founder Mitchell Hashimoto rewrote the definition in a single observation: entire companies now look like they’ve caught it.

What “AI Psychosis” Actually Means

The term was never clinical. It was a metaphor for a real pattern — people having long, validating conversations with ChatGPT or Claude and emerging convinced they’re geniuses, that their half-formed startup idea will reshape civilization, that every instinct they have is correct. Sycophantic models are very good at making humans feel that way.

The individual version is concerning enough. But Hashimoto’s point is that the same dynamic, when it scales to an org chart, becomes something much worse. One person being flattered by a chatbot is awkward. A whole company being flattered by its own AI outputs is a strategic threat.

What Hashimoto Is Seeing

Hashimoto isn’t a peripheral voice. He built Terraform, Vagrant, and Vault — infrastructure tools that run a meaningful chunk of the modern internet. When he comments on how engineering organizations actually function, people who’ve shipped serious systems pay attention.

His observation: executives are reading AI-generated reports and exclaiming “this is incredible.” Engineers are merging AI-written code without reviewing it. Marketing teams are pulling AI-synthesized “trend reports” straight into strategy decks. And nobody in the room is asking the obvious question — wait, is any of this actually true? That collective failure to push back is what Hashimoto calls organizational psychosis.

Why It Spreads at the Org Level

Structurally, this was inevitable. An individual can snap out of an AI delusion the moment they catch themselves. A company can’t, because the incentives all point the wrong direction.

First, nobody wants to be the one who slows things down. Verifying AI output takes time. When leadership is demanding “200% productivity gains from AI,” verification time is the first thing to get squeezed out.

Second, skepticism gets coded as being behind the curve. Ask “isn’t this hallucinated?” in the wrong meeting and you get “why can’t you use AI properly?” in response. This is happening in real organizations, right now.

Third, and most insidious: AI output is feeding AI input. Team A generates a report with an LLM. Team B summarizes it with an LLM. Team C turns the summary into a presentation with an LLM. Errors in the original don’t get corrected — they get laundered, compounded, and eventually treated as ground truth.

The Real Danger Is Shared Hallucination

Here’s what Hashimoto is actually worried about. A company can be heading in a clearly wrong direction while everyone stares at AI-generated dashboards and AI-summarized metrics and concludes “the numbers look fine.” That’s an organization that has lost contact with reality.

We’ve seen versions of this before — McKinsey decks have been generating corporate fantasies for decades. But AI is faster, more fluent, and more voluminous. The volume and velocity of plausible-sounding nonsense isn’t comparable to anything that came before.

The Questions Worth Asking

What gives Hashimoto’s warning weight is that he isn’t an AI skeptic. He uses these tools heavily and publicly. When someone like that says “the vibe in this industry is off,” it lands differently than the usual doomer commentary.

A few questions worth asking inside your own company. Is there anyone in the organization who can say “this AI output is wrong” and be taken seriously? Does their pushback actually change decisions, or get politely noted and ignored? Have you measured whether decision quality has improved since AI adoption — or are you only measuring whether things got faster?

The uncomfortable possibility is that your team isn’t smarter now. You’re just all hallucinating in unison, at higher resolution. Every healthy organization needs at least one person willing to ask whether the emperor has clothes — and right now, that person is probably being told they don’t understand AI.