Hashimoto coined the term 'AI psychosis' to describe entire companies operating under a shared delusion that AI will transform everything immediately, without measurement. His framing is deliberately clinical — implying a genuine break from rational decision-making, not mere enthusiasm or over-investment.
The editorial argues that companies are ripping out working systems to replace them with AI alternatives that perform worse and cost more, reorganizing teams around AI initiatives with no success metrics, and treating engineers who raise reliability concerns as blockers rather than quality advocates.
The editorial emphasizes that Hashimoto built Terraform, Vagrant, Vault, and Consul used by millions, shipped Ghostty in Zig after leaving HashiCorp, and has a decades-long track record of correct technical bets. When someone with this shipping record says the industry has lost the plot, it warrants serious attention.
The post garnered 1,408 points and 701 comments on Hacker News, suggesting the term 'AI psychosis' resonated as catharsis for thousands of engineers watching their organizations make decisions they consider irrational. The score indicates not just agreement but relief at having a credible source name the pattern.
Mitchell Hashimoto — the engineer behind Terraform, Vagrant, Vault, and Consul, and co-founder of HashiCorp (acquired by IBM in 2024 for $6.4B) — posted a blunt assessment on Twitter: he strongly believes there are entire companies now operating under what he calls "AI psychosis." The post immediately went viral, racking up over 1,400 points on Hacker News and triggering one of the most active discussion threads of the year.
Hashimoto isn't a random contrarian. He built infrastructure tools used by millions of developers, shipped Ghostty (a high-performance terminal emulator written in Zig) after leaving HashiCorp, and has a decades-long track record of making correct technical bets. When someone with this caliber of shipping record says the industry has lost the plot, it's worth paying attention.
The term "AI psychosis" is deliberately clinical. It implies not just enthusiasm or over-investment, but a genuine break from rational decision-making — companies making choices that don't connect to observable reality, driven by a shared delusion that AI will transform everything it touches, immediately, without measurement.
The Hacker News score alone tells you something. A 1,400+ point post isn't just agreement — it's catharsis. Thousands of engineers are watching their organizations make decisions they consider irrational, and they finally have a phrase from a credible source that names the pattern.
Here's what "AI psychosis" looks like in practice. Companies are ripping out working systems to replace them with AI-powered alternatives that perform worse, cost more, and introduce new failure modes — but look better in board decks. Teams are being reorganized around AI initiatives with no clear success metrics. Hiring freezes are justified by claiming "AI will handle it" before any evidence supports that claim. Engineers who raise concerns about reliability, latency, or cost are treated as blockers rather than the quality gates they actually are.
This isn't an argument that AI is useless. Hashimoto himself has been building with AI tools — Ghostty's development involved AI-assisted coding workflows. The distinction he's drawing is between measured adoption ("this tool saves us 2 hours per PR review cycle, here's the data") and psychotic adoption ("we're an AI-first company now, rebuild everything").
The pattern has historical precedent. The blockchain era produced the same symptoms: companies bolting distributed ledgers onto problems that needed a PostgreSQL table, reorganizing around "Web3 strategy," and firing skeptics. The dot-com bubble before that. The specific technology changes; the organizational psychosis is a recurring bug in how companies process hype cycles.
What makes the AI version more dangerous is that AI *actually works* for many use cases. Blockchain hype was relatively easy to debunk because the technology genuinely didn't fit most applications. AI code generation, summarization, and analysis provide real value in specific contexts. That partial validity makes the psychosis harder to diagnose — it's mixed in with legitimate improvement, and leadership can always point to the wins while ignoring the losses.
The Hacker News discussion revealed a clear split. On one side: senior engineers and engineering managers reporting exactly the patterns Hashimoto described. Teams forced to integrate AI into products where it adds no value. Performance reviews tied to "AI adoption metrics." Working prototypes shelved in favor of AI-powered alternatives that are still in research phase.
On the other side: engineers at AI-native companies and startups arguing that the skeptics are the ones disconnected from reality. Their argument: AI capabilities are genuinely improving at an unprecedented rate, and companies that don't aggressively integrate will be left behind. The cost of being too slow is higher than the cost of some wasted experiments.
Both sides are partially right, which is exactly what makes organizational decision-making so difficult right now. The engineers saying "this specific AI migration is a disaster" are usually correct about the specific case. The strategists saying "AI will be table stakes in 18 months" are probably correct about the trend. The psychosis emerges when the trend argument is used to override specific engineering judgment — when "AI is the future" becomes a thought-terminating cliché that shuts down any discussion about whether *this particular application* makes sense *right now*.
Several commenters drew parallels to the Agile adoption wave of the 2010s, where a set of genuinely useful practices got distorted into a compliance regime that made teams slower. The pattern: useful technique → management discovers technique → technique becomes mandatory → original value is destroyed by cargo-cult implementation. AI is somewhere between steps 2 and 3 at most organizations.
If you're an IC engineer watching your organization make decisions that feel disconnected from technical reality, Hashimoto's post at least gives you vocabulary. But vocabulary doesn't ship. Here's what actually helps:
Measure before and after. If your team is replacing a system with an AI-powered alternative, establish baseline metrics *before* the migration: latency, error rate, cost per request, developer time spent on maintenance. The single most effective antidote to AI psychosis is a spreadsheet with real numbers. Most irrational AI adoptions don't survive contact with a well-instrumented A/B test.
Distinguish between "AI-assisted" and "AI-replaced." AI code review that flags potential bugs for a human reviewer? Measurably useful. AI code review that auto-merges without human oversight? A different risk profile entirely. The psychosis typically manifests in the jump from augmentation to replacement without evidence that the replacement actually works.
Protect your optionality. If you're being asked to architect around an AI service, build the abstraction layer that lets you swap it out. The AI provider landscape is shifting fast enough that any tight coupling is technical debt, regardless of whether the AI itself works.
Be the person who asks "compared to what?" Every AI-powered solution has an alternative: the existing system, a simpler heuristic, a rules engine, a human workflow. The psychosis relies on never making that comparison explicit.
Hashimoto's "AI psychosis" framing will probably age well — not because AI will fail, but because the current wave of irrational adoption will produce enough visible failures to force a correction. The companies that emerge strongest will be the ones that adopted AI where it demonstrably worked and resisted the pressure to adopt it everywhere else. The correction won't kill AI; it'll kill the idea that AI is a substitute for engineering judgment. That's a lesson the industry has to relearn every hype cycle, and Hashimoto just rang the bell for this one.
I feel in a really weird position where I both really dislike what AI is doing to the experience and practice of writing code, to the point where I want a job doing literally anything else besides using the computer, but also think that these tools are extremely powerful and only getting better.I th
Maybe this is what will turn software engineering into an Engineering field.Right know, prompters are setting up whole company infrastructure. I personally know one. He migrated the companies database to a newer Postgres version. He was successful in the end, but I was gnawing my teeth when he descr
Recently I had a request come through to allow finance analysts to vibe code their apps. During a discussion one of the finance managers let the cat out of the bag. Turns out our CFO had met fellow CFOs at a get together. They talked about how each of them were using AI. Our CFO was lagging behind a
I think AI rescue consulting is going to be come a significant mode of high value consulting, similar to specialists who come in to try and deal with a security breach or do data recovery.Purely AI written systems will scale to a point of complexity that no human can ever understand and the defect c
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
I’m at a FAANG and we have $300/day token quota. Personally I don’t use that much of it but management is pushing really hard for it. “the quota has been raised for a reason, use it”. Any task: “have you tried working on it with Claude?”. Every meeting “now engineer x and y will show you what h