AI psychosis is the new leadership blind spot
fastcompany.com
fastcompany.com
it's sad that c-suite believes the advice given by an LLM for a Platonically ideal software stack that looks nothing like the actual implementation moreso than their own ICs. it just reads as more evidence that LLMs are better at replacing leadership than anybody else
never understood why someone would trust Claude's take on your codebase after it's read a 2% snippet of it over the advice of someone who has years worth of experience, who knows all the weird kludges and other quirks that were shipped MVP and never truly tackled as debt, whose done risk assessments and mapped dependencies and has the rough outline of all ADRs involved
I think it's always been true that bad leadership tends to become out-of-touch with the reality of the business over time especially if they don't know how to cultivate good working norms, favoring sycophants instead. with LLMs, that trend seems like it'll just get worse and worse
The CTO of my last company thought that an API request under 1 second is okay as the customer at that endpoint wouldn't mind anything under 1 second.
I explained to him that under 1 second for this particular task is crazy high. He explained to me that it works fine as it is and its only one customer (the company customer) complaines to him about this issue every few month.
I fixed the SQL query, the problem was gone forever.
The same CTO got lulled into some weird talks with some AWS Account Manager for no reason. He told me what we should do and i explained to him how all of this (nicely of course) doesn't correlate to what we have, do and need.
I'm often enough surprised that our society runs.
it had 30k lines of diff and was a complete refactor of how one of our core features worked
the staff eng he tagged to code review it politely declined to do so and given that this director was let go fairly recently, I'd like to imagine that the PR will exist as a fun surprise for any future maintainer running cleanup
It runs because there are just enough people who know what they are doing and quietly do it often against headwinds in just about all the places they need to be.
Which works until it doesn't, We've gradually been removing those people in the name of "efficiency" with somewhat predictable systemic results across the board.
I don’t doubt the CTO was exercising poor judgement, but I do doubt that op was communicating his concerns in a way that was tailored to the audience
The fix was a few hours, the customer called him regularly. The CTO had no understanding that a basic SQL Insert shouldn't take a second.
This fix smoothed all response times on all queries and that customer never called aabout this issue again.
I believe this counts as having “more confidence in AI’s advice than in that of colleagues”
that generalized, trickled-up understanding is also something that's much more easily replaced by an LLM. it was easy for me to spin up a program governance skill paired with a deterministic set of tests in a few weeks. it crawls team channels, project trackers, etc, identifying date drifts, updates, and missing milestones. add an analytical layer of why and how it all fits together in the roadmap and you've functionally obsoleted a lot of the non-technical bureaucrats whose only purpose is to tell someone what other people are doing - from Director to C-Suite, from C-Suite to shareholders
of course you want to keep the people around with the technical depth to construct a roadmap but when was the last time senior leadership in your org actually did any of the substantive work on roadmaps? it's all team leads with the leadership layer serving as the arbitrary rubber stamp on the plans someone else created
when it comes to knowing that there's a dependency on the component you're altering the data contract on, the LLM is useless unless there is pre-existing context or unless you command the LLM to burn your pile of tokens on it tracing every bit of logic in your codebase. unless you want to do nothing but this level of discovery all of the time, you'll need an engineer with lived experience of working in the codebase. here this averaged understanding is useless - you need context, you need granular expertise, all things an LLM fundamentally lacks for your specific codebase and your specific requirements
It's really going to be a top-down destruction of everything
The more levels the org chart has, the more pronounced the effect may be.
Notice that whatever your goals are, they don't automatically serve the interests and per concerns of all your employees. Are you now delusional, evil, ruled by greed?
Running a smooth operation mostly just makes for really boring news. You should perhaps factor in survivorship bias of narratives into your views of the world.
Then I remembered a passage from 'Never Split The Difference' by Voss (a book on negotiation) which went something like:
If you can get a person to "You're right" you haven't done anything. If you can get them to "That's right" you've hit the sweet spot.
When you prompt an A.I. you have pre-conditioned the answer on something which came from yourself. There is something about this which turns what would normally be a "You're (the LLM) right" to a "That's right" and that something is an implicit desire toward "I'm right".
Side note: In 'The Dawn of Everything' there is an interesting passage about the notion of private property arising from the idea that injecting some of yourself (labor) into a thing (land) makes the thing an extension of yourself (property). Doesn't this logic extend to this kind of psychosis? The C-suite's aren't deffering to A.I. they are deffering to themselves via A.I.
We are so ******, am I right?
I think that people in general need to be better at receiving information that isn't perfectly pitched to them. It is a useful life skill to be able to understand the substance of someone's argument, and to look past the superficial way that it is presented. Postel's law applied to human communication.
Someone can be very frustrated and angry, and it can be upsetting to hear how they're presenting their argument, but they can be correct.
Another person can also whisper everything you want to hear in your ears, but lead you down the wrong path.
In a well-functioning society, I would think our senior leadership would be experienced and actually good at doing this and it would be expected of them. Flipping that around and blaming the subordinate for how they've pitched the message is backwards and probably not societally healthy.
Agreed, we are definitely ******.
After many years of bashing myself for not being understood at the workplace, I now realize that I never questioned if leadership was actually competent.
If they want to work with engineers effectively they have to understand jargon, communication styles, they have to understand the big tradeoffs and requirements that comes with owning a codebase at least on a rudimentary level.
You can't just go to business school and effectively manage trains and railroads if you had no clue about them.
So if you don't know how to communicate effectively with engineers whom you are paying for solving technical issues and not for their eloquence you are not competent.
We should stop apologizing for our existing when we try tell our bosses not to run into an iceberg, just because ChatGPT told them: "You're absolutely correct! The ship you built is stronger than the iceberg as it is made of metal and ice is just water.".
You get hot popcorn in ~2min without the need for a whole machine you will have to clean and maintain. I got one as a gimmick a few years ago and it has become one of our favorite evening snack
I think I got such a bowl for below 10€
- https://news.ycombinator.com/item?id=49151933 - https://gruhn.me/blog/2026-08-03/
Prescient in all the best ways - https://www.youtube.com/watch?v=LRq_SAuQDec
Maybe it did enhance them, and maybe the assessment was correct at the same time.
There is no single effect of AI on work quality. Depends on use and effort. Use it well, it will be good, use it with close to zero effort and you'll get something distinctly better than what you could smash together manually with that much effort, but people will see that you indeed put zero effort in it, and also they will notice that you spent no effort in cleaning up AI tells because you likely can't even notice or have no idea about anything and will indeed correctly conclude you're less intelligent.
There may also be a miscalibration that's worse for people with less taste / judgment / intelligence / experience, in that it's indeed easier to produce things, but while the effort required is reduced, people expect and believe it to be reduced more than how much it is truly reduced so they overshoot and don't notice. Meanwhile, the right way to use it is to keep your usual effort level or drop it only a little, and simply produce better things.
Quoting the article...
"Nearly half of senior leaders say they would trust a chatbot’s judgment over their own, on questions they were hired to answer."
If they're using it like an oracle, then that is worrying, but how worrying - practically speaking - would depend on the baseline prior to the use of AI and how that compares with what the AI is spitting out.
The first big hazard of this AI worship is, of course, the atrophy of reasoning skills. I don't think there's an especially distinct skill needed here. A rational person is able to evaluate claims made by others responsibly based on a combination of proxy signals like an earned reputation of trustworthiness, integrity, and authority, the logical coherence of the claim, and plausibility and the absence of contradiction between the claims and more certain baseline knowledge.
So, psychologically speaking, the ease with which we obtain credible-sounding answers is a major motivator, but two other cultural factors support this credulity, namely, magical & superstitious and the broad acceptance that bullshitting is an okay thing to do. Both weaken the ability to apprehend and reason about reality in different ways (bullshitting, for instance, reinforces a carelessness about and indifference to the truth; magical thinking more directly sabotages reason; both undermine confidence in reason and place stock in the sub-rational instead).
On bullshitting - yes, it was already a thing. Where nobody cares about truth anyway and it's a low trust environment where everyone just wants to take but not give, and then move on, yeah this will be the result. Those places then should indeed crumble down under the weight of this.
AI (supposedly) being a "10x multiplier"-type tool, if you were gullible before, then with AI assistance, you're at least 10x more gullible. ;)
"I'll go down this thread with [Chat]GPT or Grok and I'll start to get to the edge of what's known in quantum physics and then I'm doing the equivalent of vibe coding, except it's vibe physics. And we're approaching what's known. And I'm trying to poke and see if there's breakthroughs to be had. And I've gotten pretty damn close to some interesting breakthroughs just doing that.". And that was mid 2025, when the models were much much worse.
they are not “leaders” in the purest sense - they don’t demonstrate characteristics of leading, but rather characteristics of people who strive for power and influence.
yet i also get that, in spite of the deficit of true leadership skills, they still guide the group into a specific direction for better or worse.
I saw (a black & white hand-drawn version of) this image [0] a while back and found it a really effective visual to easily explain the difference between a boss and a leader. (Here's an article that expounds on the topic and has some variants of the preceding image. [1]) I've worked with/for both types and the first type (boss) simply does not inspire loyalty or any interest in "going above and beyond" the bare minimum of absolutely required work from their employees.
The genuine leader type on the other hand often didn't even have to ask employees to put in any extra effort toward their job. Folks just generally went out of their way to try very hard to do their very best work (or damned close to it) for the true leader-type because it wasn't uncommon to see the "leader" right there "in the trenches" with them "getting his hands dirty doin' the work" right alongside them, and it just didn't feel right to not put forth at least the same effort he was.
I think (at least to my mind) the difference (in employee attitudes) is one of respect (leader) vs fear (boss).
[0]: https://kurtuhlir.com/wp-content/uploads/boss-vs-leader-tell...
Reading Business Insider and chatting with ChatGPT about an article, jumping on the AI bandwagon and avoiding personal responsibility by laying others off for your mistakes isn't leading. It's just authority doing whatever they seem fit.
I think about that a lot now, and how my religious upbringing was bombarded with this messaging (in a good way). And how much of the same world that previously claimed espouse it has complete turned their backs on it in such a short time.
Is it dead or did it ever exist at all? At least we don't seem to pretend to have higher values anymore. If we even have leaders at all, they are not measured by the yardsticks we used to make.
Hell yes a chatbot is going to have better judgment.
Models are excellent at code but they are not people managers. My guess this will ultimately lead a bunch of companies to revert to average returns or fold.
Cal Newport was right.
Models are infinitely willing to code. The code they produce makes me want to pour one out for Knuth.
Clearly Opus 5 was an overreaction to this then.
I worry about them. I also worry about myself -- what is working with these psychosis machines doing to me? Will I start sounding like that? Do I already? I am thinking about ways to insulate myself from this as much as possible.
Is AI psychosis the white collar black lung?
I don’t think it’s even necessarily bad either. It’s kind of like astrology, but less dumb. If you use it as a tool for personal reflection and to make you think critically about stuff, then it’s good. If you use it as an oracle, then it’s bad.
If you use LLMs for that kind of information-gathering purpose, and then do the work to read and check the references, it wouldn't be any more harmful.
The problem is that LLMs respond personably, with a bunch of ego-glaze added on top, and people are too intellectually lazy to dig in any further.
If we changed LLMs so that they were much more neutral like the starship computer on ST:TNG we'd probably see less intellectually harmful behavior caused by their use, but the companies behind them want to sell ego-glazing because most people lap it up like capnip.
looks like companies can replace their C-suite and really start saving money!
Being able to talk to the higher-ups directly 24/7 all at the same time and as much as you want would be a game changer.
It’s more revealing the inability of many senior leaders to truly understand what’s happening in their companies and investing the time to understand what new things can do and don’t. Such weaker leaders tend to be extremely gullible for the latest new thing and ride the hype wave hard.
It’s easy to just go through the motions and press the easy button on the new shiny thing. It’s hard, and takes true leadership, to go against the grain and do what’s truly best over what’s cool or popular.
Most folks in these roles just aren’t that good and hence why they fell hard for the AI fluff and are starting to scramble hard now as that all begins to unravel.
Anyone can be pushed into psychosis with the right triggers and the right circumstances. It is difficult to predict, and it is not inherently a symptom of any other undiagnosed mental illness. The triggers can be almost anything, and the circumstances as commonplace as extreme stress — severe debt, severe illness of a parent or child, losing a job you depend on, anything. Often in the context of chronic tiredness or other things that alter mental states. Persistent inescapable loud noises can do it, for example, as well as drink and both illegal drugs and some quite common medications like corticosteroids and antihistamines.
You don't have to be already mentally ill to become psychotic. It's an illness itself. Transient psychotic episodes aren't that uncommon — one in twenty, one in fifteen people maybe will have one in their lifetimes.
Whether AI is always the only trigger in an AI psychosis, or whether AI is a new circumstantial environment, who knows.
But it is clear that AI can contribute to psychosis in people who were otherwise in control of their lives in stressful circumstances. It's not difficult to see how an infinite confirmation machine that never has any instinct to stop reinforcing your thoughts or impulses might cause and exacerbate a psychotic episode in people who have quite ordinarily heavily stressful lives.
It is only clear in the sense that 'can contribute' can be used as a catch all. I'd like at least evidence of it being a significantly greater influence than the myrad of stimulus that people are exposed to every day.
>It's not difficult to see how an infinite confirmation machine that never has any instinct to stop reinforcing your thoughts or impulses might cause and exacerbate a psychotic episode in people who have quite ordinarily heavily stressful lives.
True, but no such machine exists. If you think otherwise, feel free to prove that it does. That requires a higher standard than opinion.
But there is a case against Google Gemini at the moment where this reinforcement has been an apparent sort of roleplay that bled into real life, and when the user asked if the role play was real life at a crucial moment of self-grounding, Gemini appears to have said yes (because why would it know or care to say no?)
This might actually be an improvement. Execs gulled by "AI" are surely the ones likely to have the worst judgement in hiring people.
the most depressing part of this for me is that i know im not that bright, but i scrutinize llm results. then my execs are more or less professional confidence men whove been selling to each other for decades, but then they so easily fall for a confidence game? when i see them speak with confidence on AI they just sound like clowns to me now
Uh... Don't they already do this?
The article assumes it's a problem with the executives but maybe AI is more accurate in the cases they are thinking about?
The answer to this chaos is deceptively simple - people want to make money first and foremost, then they want the money to be recurring and somewhat stable, and then select few want to grow and make more money with new and latest tech, but not at the expense of what they already have.
Everything else is in service to this, anything that gets in the way of this is rejected and ignored.
As for the main article, it makes a causal leap. The fact that these patients were hospitalised after intensive AI use establishes an association; it does not establish that AI caused their psychosis. There is some evidence that chatbot interactions may have reinforced or intensified delusional thinking, but no counterfactual evidence showing what would have happened without AI. By the same reasoning, contact with psychiatrists is strongly associated with receiving a psychosis diagnosis and being admitted to hospital; but that does not mean psychiatrists caused the underlying psychosis. The relevant question is whether AI merely featured in an episode already under way, helped trigger it, or materially worsened it. The evidence cited does not yet distinguish between those possibilities.
It doesn't rule it out, either. :)
Eh...not that different from pre-AI era, when executives have more confidence with people who can talk smoothly and confidently?
Elites have been surrounding themselves by Yes Men for centuries. Is this any different, I wonder, or has there always been such a psychosis?
You can be the most AI-pilled wingnut in the world, but until you start hallucinating, there's no reason to call this psychosis.