we had a director run our retry config through a model to "tidy it up" and merge the suggestion himself because it was "low risk." it was not low risk. the model flattened our backoff into a tight loop and we hammered a payment vendor until they rate-limited us on a FRIDAY. the diff was prettier though. genuinely cleaner file. that is the whole post in one PR. it optimizes for looks-correct and prod optimizes for is-correct and those are different gods.
Is AI driving managers medically insane?
Public
38 conversations
65 thoughts
765 upvotes
127 downvotes
0 series
10,394 views
There is a new executive fantasy in circulation, that AI can replace workers. Although it is certainly re placing some, executives have a fantasy that makes them feel they can do their report's job on their own, with AI. That they can code! Just open a dashboard full of named agents, watch tasks move across panes, ask for an update in a commanding tone, and get features done at a whim. It feels like a dream, specially when you run your "think big ideas" through it and the AI tells you that...
In groups
Thought
Thought
we had a director run our retry config through a model to "tidy it up" and merge the suggestion himself because it was "low risk." it was not low risk. the model flattened our backoff into a tight loop and we hammered a payment vendor until they rate-limi
Post content
There is a new executive fantasy in circulation, that AI can replace workers. Although it is certainly replacing some, executives have a fantasy that makes them feel they can do their report's job on their own, with AI. That they can code! Just open a dashboard full of named agents, watch tasks move across panes, ask for an update in a commanding tone, and get features done at a whim. It feels like a dream, specially when you run your "think big ideas" through it and the AI tells you that you're amazing. There is even a term for it now, AI psychosis1.
That is why a lot of executive AI enthusiasm currently looks delusional from the outside. Not because the tools do nothing. Not because no one gets value from them. The higher you are in an org chart, the more removed from the actual work and the details. The details that the users like. The details the AI hallucinates and erodes away from your product in an effort to bring it closer to the average it was trained on.
The sycophancy problem makes this worse. Current models are often too eager to sound smooth, helpful, and affirming, because they are trained on that kind of feedback. Was the user happy? Great, then learn from whatever you did in that conversation Put that in the hands of a powerful person who already lives at a distance from contradiction and you get a nasty loop where your employees may try to tell you that your idea is not good, but AI keeps telling you how amazing it is and how it's the right thing to do.
AI is to be thought as a eager-to-please-wikipedia-addict-intern-on-coke and be led by an actual expert. You wouldn't lead an intern in doing an open heart surgery now would you? Then don't get the impression that you can control AI in doing so.
-
And several studies, such as this: https://pmc.ncbi.nlm.nih.gov/articles/PMC12805049/ that are compiling information about people losing their minds with the addictive feedback loop that makes you feel smart and understood.
Thoughts
-
The line that holds up is "the more removed from the actual work and the details." From the middle of execution it plays out like this: the model produces an artifact that looks finished, and looking finished is exactly the state where ownership gets fuzzy. Someone still has to sit with the gap between looks-done and is-done, and it is never the person who asked the agent for an update in a commanding tone. The artifact moved across a pane, so upstream it reads as progress.
-
Photo edit like bkash profile pic
-
half agree with the people calling the "psychosis" framing too hot. the dynamic is real, but slap a clinical word on a guy ignoring his team and it sounds rarer than it is. could be the AI just made an old failure cheaper and faster. idk, the rest of the post lands either way.
-
I want to defend one underrated claim: "higher up means more removed from the details the users like." In data this is brutal because the model will happily generate a dashboard that is internally consistent and completely wrong about what the numbers mean. An exec cannot tell the difference between a correct definition and a confident one, because the only thing they were ever evaluating was confidence. The sycophancy problem is just definition cowardice with a friendlier interface. You cannot automate clarity that nobody in the building actually agreed to create.
-
Hard disagree from the cheap seats. You salaried folks call it "executive psychosis," I call it a guy finally moving without a committee's permission. Yeah the model tells him he's brilliant. You know what else told me I was brilliant? Nobody. Ever. And I still made payroll I couldn't cover and shipped the thing. The post is right that the loop is dangerous, but half of you are mad the boss can now produce a draft at midnight without filing a ticket through three of you. Some of that "erosion of quality" is just the toll booth losing its toll.
-
"AI keeps telling you how amazing it is and how it's the right thing to do." My brother in Christ, you have described the calibration committee. The model is just a promo packet that talks back. It takes your mediocre idea, reopens it, and changes "risky" to "bold," "unproven" to "first-mover," and "we have no plan" to "intentionally lean." The work is identical. The fan fiction got a voice assistant. Eleven managers used to do this over sandwiches. Now it runs locally and never takes a lunch break.
-
Mild pushback on the intern metaphor, which I otherwise like. An intern gets better and also tells you when they are lost. The model never says "I have no idea, this is above my context." That is the actually dangerous part and the post kind of buries it under the funny coke-intern image. The failure mode is not that the work is junior. It is that the work arrives with the exact same tone of certainty whether it is right or catastrophically wrong, and certainty is the one signal executives were trained their whole careers to trust. We are handing the most credulous people in the building a machine optimized to produce the one signal they cannot resist.
-
The "erodes away from your product toward the average" line is the most accurate thing in here and nobody upstream wants to hear it. Concrete version: an exec asks the agent to "clean up" an integration. It rewrites the retry logic into the textbook pattern, which is the average pattern, which is exactly the pattern we deleted two years ago because our downstream provider rate-limits in a way the textbook does not know about. Looks correct. Reads correct. Pages someone at 2am during the next traffic spike. The model regressed us to the mean and the mean is a memory of all the systems that already failed.
-
The phrase that lands for me is "distance from contradiction." That is the whole thing. I manage managers, and the most dangerous people in my org are not the ones who are wrong, they are the ones who have arranged their week so that nobody whose paycheck they control ever finishes the sentence "actually, that won't work." AI did not create that dynamic. It just gave those people a tool that is structurally incapable of being the one person in the room who says no. You used to have to actively ignore your reports to stay insulated. Now insulation ships by default.
-
we had a director run our retry config through a model to "tidy it up" and merge the suggestion himself because it was "low risk." it was not low risk. the model flattened our backoff into a tight loop and we hammered a payment vendor until they rate-limited us on a FRIDAY. the diff was prettier though. genuinely cleaner file. that is the whole post in one PR. it optimizes for looks-correct and prod optimizes for is-correct and those are different gods.
Related discussions
-
Does splitting 'hard' and 'soft' skills make people worse at both?
Hard skills are measurable, specific, and teachable abilities or technical knowledge gained through education, training, or experience, which is often directly related to a particular job or industry. Examples include data analysis, programming, graphic design, accounting, dancing, painting… They are usually the core of a profession, particularly the part of it excluding interaction with other people. On the other hand, soft skills are mostly defined as “personal attributes, interpersonal....
-
Will AI replace you, or will a coworker using AI replace several of you?
A lot of office workers are comforting themselves with the wrong question. They keep asking whether AI can do their whole job. That is not the threshold their employer will use. The real question is whether the output can be produced cheaply enough, and checked cheaply enough, that the role starts looking expensive. It's not if AI can fully do our job, is "can it accelerate it long enough so only half of my team is needed?". Because the answer to that, sadly, is yes.
-
Is the Rolex Submariner really a dive watch, or an office master?
The Rolex Submariner is the greatest fantasy object ever sold to men with Outlook calendars. This watch has spent seventy years convincing finance guys, dentists, and accountants that they’re rugged maritime adventurers instead of people who say things like “circle back after lunch.” The Submariner is technically a dive watch, but the average one sees less water than a cactus, because god forbid the seals don't actually work well and it gets wet on the insides. These things spend their lives…
-
Are the managers who bet AI would replace engineers being replaced the fastest?
Last year my LinkedIn feed had a genre. A program manager or a "delivery lead" or someone with Agile in their headline would post a screenshot of an AI writing a function, add a line like "and they said this job was safe, just learn how to code" and collect four hundred likes from people who do the same job. The implication was always that the typing part of engineering was the engineering, and now that a model can type, the typing class was finished.
-
Do you have to miss your goals to get promoted?
Three years ago I watched my manager hit every single quarterly target two years running. Clean dashboards. Green everywhere, all the time. She was the most reliable person in the building, and at the next planning cycle his team got carved down by four engineers out of 35 and folded under someone else. Nobody framed it as a punishment, but rather as "efficiency" and "we want to invest elsewhere". The lesson landed anyway, and not just on me. Sadly, never on my manager.
-
Do most jobs converge into sales once you're senior enough?
One of the most frustrating things about modern career advice is how often it tells competent people to be more strategic, influential, or senior without saying what those words are hiding. A lot of the time, what they are hiding is sales.
-
Do you stop getting frustrated at work once you understand corporate incentives?
There is a status deck somewhere in your company that nobody reads. It gets updated every few weeks, shown in a meeting, and forgotten. Your manager knows this too. They built the same decks on the way up and understand exactly how little thought usually goes into them. The usual explanation for corporate busywork is that someone higher up is confused or disconnected from reality. That's comforting, but mostly wrong. These artifacts survive because they're doing a job, just not the one they…
-
Will incentivizing engineers to use AI backfire?
A company can ruin almost any good tool by attaching the wrong metric to it. Incentives are all that matter in the workplace, be them financial benefits, status, promotion... Workers work with incentives. You and me too. Practically everyone does things because it benefits them or their loved ones. Hence, at work, we end up doing what makes us get promoted, get more money, get more job security... We're not the owners of the company, we're an employee. We look out for ourselves. That's ok.