A software engineer with a long public track record posted something blunt on Hacker News this week: he believes entire companies are currently operating under what he called "AI psychosis" — a state where leadership has so thoroughly outsourced judgment to AI tools that the organization has lost its ability to distinguish confident-sounding output from correct output. The thread drew thousands of responses, most of them from people saying: yes, I work in one of those places.
That's a signal worth sitting with.
What's actually changing
The problem isn't that companies are using AI. The problem is a specific failure mode: organizations where AI output has become the default source of truth, where the humans nominally reviewing that output no longer have the domain knowledge — or the organizational permission — to push back.
This creates compounding risk. An AI system confidently misidentifies a market, a cost structure, a regulatory requirement, or a personnel decision. A manager passes it up the chain because it sounds rigorous. Leadership acts on it because it arrived in a polished format. Nobody in the loop has either the expertise or the incentive to say "wait."
The consequences are not abstract. Hiring freezes based on AI-generated headcount models. Product pivots based on AI-synthesized competitive analysis that missed key context. Layoffs structured around AI-recommended efficiency ratios. Contracts awarded or canceled based on vendor assessments the AI got directionally wrong.
For workers, this means a new category of career risk: your job security is now partly a function of whether your employer's AI pipeline is trustworthy — and you have essentially no visibility into that.
Recent data on white-collar layoffs suggests that mid-level roles in strategy, research, operations, and middle management are disappearing faster than the broader labor market would predict from economic conditions alone. The Hacker News thread is one data point, but it rhymes with what BLS data on occupational displacement has been showing for several quarters: the cuts are concentrated in exactly the roles that AI systems are confidently, if not always correctly, replicating.
The secondary risk is subtler. When a company operates on bad AI outputs long enough, it tends to underperform, lose clients, or make a consequential strategic error. That produces a real business contraction — layoffs, restructuring, or closure — that feels to workers like a normal downturn but was actually downstream of a decision-making failure that happened 12 to 18 months earlier. By the time you see it coming, you're already in it.
What we'd actually do
Build a paper trail of your actual value, now. Start documenting concrete outcomes you've produced — revenue influenced, costs reduced, problems solved — in your own records, not just in your employer's systems. If your company's AI-assisted performance reviews no longer reflect reality, you need a private archive you own. A simple running document, updated monthly, is enough.
The goal is interview-readiness that doesn't depend on your manager's memory or a software system's summary of your tenure. If your company's institutional knowledge is increasingly mediated by AI, your actual contributions may be less legible to the organization than they used to be. Make them legible to the outside world instead.
Treat your industry network as infrastructure, not networking. Two to three genuine professional relationships outside your current employer are worth more right now than a polished LinkedIn profile. These are people who have seen you work, who would take your call, and who exist in a different organizational context than you. If your company enters a contraction driven by AI-compounded errors, your next opportunity is most likely to come through a person, not an application portal.
Run a household income stress test. How many months can your household operate at current expenses if your primary income stops? The honest median answer for American households is somewhere under three months. Pushing that number to five or six — through a combination of reduced fixed costs and a small liquid reserve — is the single highest-leverage preparedness action available to most families. It doesn't require predicting whether your company specifically is in AI psychosis. It insulates you from the outcome regardless.
Develop a legible secondary skill. Not a side hustle empire — one skill that produces value outside your current employer's context and that a human client can evaluate directly. Consulting, instruction, a trade skill, licensed services. The point is optionality that doesn't depend on another organization's AI pipeline deciding you're worth hiring.
The bigger picture
The Hacker News thread will be forgotten in a week. The underlying pattern won't be. Organizations are still in the early stages of learning where AI judgment fails, and the learning is happening through real failures with real consequences for the people inside them.
You don't need to predict which companies are affected. You need a household that is durable enough to absorb a disruption of six to twelve months if one hits — and flexible enough to move when the signal finally becomes clear. That has always been the goal. AI makes it more urgent, not categorically different.
