Chapter 9: The Dead Knot of Corporate Culture
AI can generate ten architectural proposals, fifty pages of a business plan, and three different product prototypes in a single hour. Technical bottlenecks are genuinely disappearing. But hand these things to a company — five people, ten people, nothing extraordinary — and the time spent arguing over "which plan to use," "why not mine," and "is this direction even correct" can easily consume a year or two.
AI has compressed build time to a matter of days, but the time humans take to reach consensus has not been compressed by a single second.
And so an exquisitely ironic situation emerges: AI has made building nearly free, but organisations have made decisions extraordinarily expensive. Before AI, build costs were high, so the slowness of decision-making was concealed — construction would take six months anyway, so spending an extra two months in discussion seemed inconsequential. Now that building takes only a few days, those two months of discussion are nakedly exposed. They are not "caution." They are waste.
The most expensive cost in any enterprise has never been technology. It is consensus.
This is not a feeling; it is measurable. A McKinsey survey found that 61% of executives say at least half the time they spend on decisions is ineffective; only 37% of respondents believe their organisation's decisions are both timely and high-quality. Executives spend an average of 23 hours per week in meetings, and 71% consider meetings unproductive. And Asana's annual survey is more direct: knowledge workers spend 58% of their day on meetings and administrative tasks — not doing the work they were hired to do. These numbers existed before AI became widespread. All AI did was make them impossible to continue ignoring.
The Battlefield of Discourse Power
Why is consensus so expensive? Because on the surface it is communication; underneath, it is power.
When AI can simultaneously generate multiple viable plans, the question "what should we do" no longer has a single correct answer. Every plan has its merits; every direction has its arguments. In this situation, decision-making is no longer a cognitive problem — "which plan is best" — but a political problem — "whose definition gets adopted."
Within a team, the tech lead believes the system should use Architecture A; the product manager believes it should use Architecture B; the marketing director believes neither matters and they should build Feature C first. All three have reasonable justifications. AI can build all three plans — A, B, and C — within a day; the technical bottleneck is zero. But the three people can spend three months arguing, because what each person defends is not just the plan itself, but their discourse power within the organisation: if my plan is rejected, does that mean my judgement is not trusted? Does that mean I will not be consulted on important matters next time?
So what you see in many companies as "technical discussions" are not actually about technology. They are power games disguised as professional dialogue. Everyone uses technical jargon, data, and frameworks to package their position, but the real driver of behaviour is not "which plan is best for the company" but "whose version gets confirmed."
AI has not only failed to solve this problem — it has exacerbated it. Because AI lowered the cost of proposing a plan, the number of plans has exploded, and every one comes with seemingly reasonable supporting arguments. More choices mean harder consensus; harder consensus means higher political costs.
The ultimate outcome is often not "the best plan is selected" but "the plan of the person with the most political capital is adopted," or "everyone compromises on a middle-ground plan that no one is satisfied with but no one opposes." Most of the plans AI generated get thrown away — not because they are bad, but because the organisation's decision-making mechanism cannot digest that many options.
The Person Who Sees the Problem
If the battle for discourse power is exhausting enough, there is an even more hidden, more cruel trap.
In any organisation, there is a type of person: because of the depth of their experience and cognition, they can see, before a project even launches, the structural problems it will hit in the future. They are not pessimists, nor contrarians — they are saying: based on my past experience, this path, after reaching a certain point, will hit a wall, and that wall cannot be bypassed with technology alone.
The landmine dismantled in Chapter 6 is an example. A person who had lived through three generations of platform transitions could, before the fourth-generation platform's beach assault, point out: the emulation layer cannot translate kernel-layer drivers; anti-cheat systems and enterprise security software will be stuck there; this is not a problem a new chip can solve. What he stated was not a prophecy but a set of constraints — he had been saying it for two years, merely waiting for history to rhyme.
But in most organisations, this type of person's position is deeply awkward.
During the project initiation phase, what the organisation needs is not a complete risk model but a narrative that pushes the project forward — vision, opportunity, market potential. These are the language that secures budget, headcount, and executive support. And constraint analysis has a very uncomfortable property: raise it too early, and everyone thinks you are too pessimistic; raise it too accurately, and everyone thinks you are obstructing progress.
Consider a scenario you have probably witnessed. A company decides to use AI to rebuild its internal workflows. At the kickoff meeting, someone points out: the existing data architecture does not support this; it is not something an API integration can fix; the foundation needs to be torn down and rebuilt, and tearing down the foundation will affect three departments' existing systems. After they finish speaking, the room is silent for a few seconds. Then the project lead says: "We'll deal with those issues later. Let's push the MVP out first." Three months later, the MVP is out. Six months later, the foundational problems surface. The cost of rebuilding is four times the original estimate. Then the post-mortem begins: "The person who raised the risk early on — why didn't they push harder for a solution?"
No one asks the other question: why didn't anyone listen at the time?
This is the mechanism of post-hoc attribution. When the outcome is bad, the organisation traces backward, finds the person who "knew all along," and asks: since you saw the problem, why didn't you solve it? Since you raised the risk, why didn't you persuade everyone? Since you have this capacity for judgement, why didn't you fight for the resources to prevent it?
Every question sounds reasonable. But every question carries the same false assumption: the person who sees the problem possesses the power to solve it.
A seismologist who predicts earthquake risk does not thereby have the ability to reinforce every building in the country. A financial analyst who identifies a bubble does not thereby have the power to change the market. An architect who flags technical debt does not thereby have the resources to rewrite the system. Between "seeing" and "solving" lies not a gap of ability, but of power.
But post-hoc attribution does not examine this gap. Its logic runs backward: bad outcome → someone knew → why didn't that person stop it. This logic translates a structural mismatch of power into an individual's failure of ability or communication.
And it has an even more concealed function: protecting the people who actually held decision-making authority. If responsibility is assigned to "the person who knew but failed to persuade everyone," then the person who made the call, the person who allocated resources, the person who chose to ignore the risk — they are all absolved.
So you see an absurd pattern: the person with the most judgement ends up bearing the most post-hoc accountability. The person who actually made the decision is always safe. Because the former is the one who "knew all along," and the latter is the one who "made a decision everyone agreed with." Knowing too much, in a corporation, is sometimes not an asset — it is a liability.
The Triple Function of Corporate Rhetoric
The costliness of consensus, the battle for discourse power, the trap of post-hoc attribution — in companies that have already been financialised, these problems are never discussed openly. Because such companies have an extremely sophisticated language system, purpose-built to translate these problems into statements that sound reasonable, even positive.
Layoffs are called "organisational optimisation." Squeezing is called "improving operational efficiency." Cutting benefits is called "flexible compensation plans." Being forced to do two people's work is called "empowerment." A culling system is called "building a high-performance culture." Every term is designed to repackage an action harmful to employees into a story that benefits everyone.
This rhetoric serves three functions.
Externally, it is PR — maintaining the corporate image. Internally, it is an anaesthetic — allowing the middle managers who execute decisions to do things they know are wrong with a clear conscience. A middle manager who must lay off half their team cannot proceed if the narrative in their head is "I'm helping the CEO save money so his options vest." But if the narrative becomes "We are undergoing a strategic transformation to build a more agile team," they can convince themselves to carry on. In court, it is body armour — every official statement has been reviewed by the legal team, designed to minimise litigation risk.
And the most masterful, most cruel move in this rhetoric is the individualisation of structural problems. If you are eliminated, it is not the system's fault — you did not try hard enough. If you cannot keep up with AI's pace, it is not a flawed transition strategy — you failed to upskill. If you are marginalised in the organisation, it is not a problem of power structure — your communication skills need improvement.
Every time, responsibility is transferred from the system to the individual. And the individual — especially those without discourse power — lacks the language, the platform, and the resources to push back.
AI will not change this rhetoric. In fact, AI may make it more refined. In the past, "organisational optimisation" still required a human to write it. Now AI can help generate an entire internally and externally consistent, logically coherent, emotionally calibrated layoff communications package — from the all-hands email to the FAQ, from the press statement to the LinkedIn announcement, all done at once. The tool has upgraded, but the logic it serves has not changed.
Institutional Self-Selection
All of this — the costliness of consensus, the battle for definitional authority, post-hoc attribution, the rhetoric system — is not an incidental phenomenon at a particular company. It is a structural feature of financialised corporate management.
Within this structure, the evaluation of managers is not "are you good to your team" but "how much did the costs in your department go down, and how much did output go up." A middle manager who is great at leading people and builds team belonging, but whose costs did not drop, receives the review comment "lacks decisiveness." Conversely, a universally disliked supervisor who is ruthless enough to push out high-salaried veteran employees, backfills with cheaper new hires, and makes the departmental budget numbers look good — their review reads "high execution capability," and they get promoted with a raise.
This system self-selects. People with conscience cannot keep doing these things and will leave on their own. Those who stay and rise tend to be those who can. Over time, the composition of management skews increasingly toward a certain personality profile — not that they are all bad people, but the system genuinely favours those who are decisive to the point of being cold-blooded in their upward mobility.
AI cannot change this selection mechanism. It can make building faster, analysis more accurate, documentation more polished. But it cannot change who holds decision-making power, who bears the consequences, whose version gets confirmed. It cannot change a company's incentive structure: the median tenure of S&P 500 CEOs is only 4.8 years (Equilar, 2022), and shrinking — by 2025 the average tenure of departing CEOs had dropped to 7.1 years (Russell Reynolds). But options are tied to quarterly performance. A CEO with a tenure of less than five years — the consequences of their decisions, good or bad, mostly surface under the next CEO's watch. It cannot change the logic of corporate power distribution: sometimes the person who knows the problem best has the fewest resources.
So the ceiling AI hits within corporations has never been a technical ceiling. It is a cultural ceiling. It is the iron triangle formed by "who gets to decide," "who gets blamed when things go wrong," and "how it gets packaged." This iron triangle cannot be trained by any model, nor automated by any agent.
But you might think: if AI truly becomes strong enough to replace a large share of middle and upper management — those people who consume time in meetings, compete for discourse power, and conduct post-hoc attribution — the bureaucratic layer gets flattened, and the problem is solved, right?
In theory, yes. But in practice, layoffs are never a precision scalpel. They are a cleaver.
When companies use AI as the justification for streamlining, what they cut is not just bureaucracy but also those who appear "not efficient enough" yet actually possess the deepest judgement. Because the criteria for layoffs are numbers — cost, output, rank — not judgement. Judgement has no KPI, no dashboard, no way to be seen on a spreadsheet. A person who has been with the company for fifteen years, witnessed three platform transitions fail, and knows which roads cannot be taken, is just one line of high salary figures on the layoff list.
And so a problem even harder to solve than the original one emerges: bureaucracy is indeed reduced, but disappearing alongside it are the only people who knew where the walls were. The structure is simplified, but the organisation's judgement capacity is diluted along with it. True efficiency has never come from fewer headcount; it comes from whether the people who remain have sufficient judgement and capability to make the right decisions. The layoff mechanism cannot discern this. It only recognises numbers.
Shuffle-and-Reprice
After the cleaver falls, the vacated positions do not stay empty. They are filled — and the manner of filling creates a distortion more covert than the layoffs themselves.
A senior manager with fifteen years of tenure and an annual salary of 1.5 million is cleared out by the retirement plan. A supervisor with five years beneath them gets "promoted" to take over, with a salary adjustment from 800,000 to 1,000,000 — a 25% raise. For that supervisor, this is a promotion, a pay increase, a major career step.
But look at what they have inherited. The project scope, decision-making authority, and cross-departmental coordination complexity that the senior manager handled are all dumped onto them. The workload is not just their old job plus the old boss's job — the entire position has been redefined after the old boss left, with management span expanded two to three times, while salary went up only 25%.
From the enterprise's perspective: a 1.5-million person has been swapped for a 1-million person, the work scope has not shrunk, and total labour cost has dropped by one-third. On the earnings report, per-capita output has improved and operating expenses have declined. The numbers are bright.
From a labour-pricing perspective: the unit price of this position dropped from 1.5 million to 1 million, a decline of over 30%. But the person in the role will not calculate it this way. Their reference frame is their own previous salary of 800,000, not the 1.5 million that the last person in this role earned. That reference frame has already vanished along with the person who was cleared out — nobody will tell them: what you are doing now was worth one and a half times the price two years ago.
This is shuffle-and-reprice. It is not a pay cut; it is a redefinition of the position's worth, then sold at a new price to a new person — who thinks they got a bargain.
And this shuffle has another layer that goes unspoken.
Those cleared out by retirement plans, cut by cost criteria, whose age plus tenure equals seventy — they share a common characteristic, not just "expensive" but also "old." Over forty-five, with the company for over fifteen years. This is not coincidence; it is the same thing: long tenure means high salary, high salary means high cost, high cost means cleared out. But throughout the entire process, the word "age" never appears on any official document. The retirement plan's criterion is "age plus tenure"; the layoff criterion is "cost" and "rank" — every term precisely avoids the legal red line of age discrimination, yet every term points to the same group of people.
Data occasionally leaks the truth. In March 2026, former Meta Senior Director Nicolas Franchet (age 54) sued the company, citing internal data Meta had provided to laid-off employees: in the February 2025 layoff round labelled as targeting "lowest performers," employees over forty were 1.5 times more likely to be laid off than those under forty; employees over fifty were 2.5 times more likely. Meta claimed it was cutting on performance; the data showed it was cutting on age. X (formerly Twitter) layoff litigation showed similar figures: 60% layoff rate for employees over fifty, nearly 75% for those over sixty, versus 54% for those under fifty. HP and Hewlett Packard Enterprise paid $18 million in 2023 to settle similar claims; Google paid $11 million in 2019. And EEOC (US Equal Employment Opportunity Commission) age-discrimination complaints rose from 11,500 in 2022 to 14,144 in 2023 to 16,223 in 2024 — accelerating for three consecutive years. An AARP survey released in early 2026 (1,656 employees age fifty and over) showed 90% reported experiencing age discrimination, and 83% felt occasionally disrespected. These numbers all say the same thing: "performance optimisation," "cost management," "organisational streamlining" — different labels, same recipients.
And those who fill their positions share an equally unspoken common trait: youth. Early thirties, five to seven years of experience, energetic, salary expectations still in the first half of the upward trajectory. EEOC data sketches the contour of this replacement: the proportion of employees over forty in the tech industry declined from 55.9% in 2014 to 52.1% in 2022, while those aged 25 to 39 expanded to 40.8% — far exceeding the national average of 33.1%. Companies will not say "we are clearing out the old." Companies say "building a young, agile, dynamic team." Same thing, different rhetoric; the nature shifts from age discrimination to organisational transformation.
This rhetoric works because it aligns perfectly with the self-narrative of the person stepping in. A thirty-two-year-old "promoted" into the position of a former forty-seven-year-old sees the narrative "the company is finally giving young people a chance," not "the company is using my lower price to replace a higher-priced person." They feel they are the one being seen, the beneficiary of this transformation. And that forty-seven-year-old — they are no longer anywhere they can be seen. Not in the office, not in industry groups, not in LinkedIn feeds. They have vanished. And vanishing itself produces no data.
So the full picture of this shuffle is: clear out the most expensive people (who happen also to be the oldest), fill in with the cheapest (who happen also to be the youngest), then package the whole thing as "team rejuvenation" and "efficiency improvement." Costs went down, average age went down, numbers improved across the board. Wall Street sees a "refreshed" company. What no one sees: the price of refreshment is zeroing out twenty years of judgement along with twenty years of salary.
Multiply this pattern by ten thousand positions, and you arrive at a deeply paradoxical landscape: the job market appears to still be functioning, people are moving, salaries are rising, LinkedIn "promotion" posts keep appearing. But the labour pricing of the entire middle layer has been systematically depressed, and everyone is looking at the wrong reference frame.
Chapter 4 noted Gartner's prediction — people laid off today will be rehired years later under new titles, "AI Coordinator," "Model Operations," only at salary grades that never return to the old level. Shuffle-and-reprice is the micro-level scene of this in action: no need to wait three years; it is happening every day, only no one is calling it a price reduction.
What the Numbers Conceal
If the story ended here, it would be merely a salary-compression problem — painful, but at least measurable. The real danger is the chemical reaction produced when this combines with the cleaver described earlier.
The cleaver cuts not just bureaucracy but also judgement. Shuffle-and-reprice fills the vacated positions with cheaper people. Combined, the enterprise gets an entirely new workforce structure: lower cost, execution capability still intact, but the organisation's judgement capacity has been diluted — the people who knew which roads not to take are gone, and those who stepped in have efficiency but lack that depth.
And the financial figures of this new structure look good. Per-capita output is up, operating expenses are down, gross margin has improved. Wall Street sees efficiency gains. The brand narrative is "AI-driven organisational streamlining." No metric is tracking the dimension of "judgement," because judgement has no KPI.
This is the hidden version of Gresham's Law.
The traditional Gresham's Law is easy to spot — bad people stay, good people leave, the company gets worse, everyone can see it. But in this version, the company improves on paper. The problem is hidden in a dimension no report can track. Bullshit jobs have indeed been partially eliminated, but eliminated alongside them is the organisation's sense of direction.
It is not visible in the short term. Because the people who stepped in have execution capability and can do the existing work well. But "doing existing work well" and "knowing what to do next" are two entirely different capabilities. Directional errors — choosing the wrong product path, underestimating technical debt, misjudging a market inflection — do not surface in the first quarter. They surface in the third or fourth quarter, disguised as a series of seemingly independent "execution problems." And by then, no one can trace it back to the fact that a cohort of people was cleared out eighteen months ago.
Even more paradoxical: this process carries a positive feedback loop. Numbers improve, the market rewards this approach. Other companies see it and follow suit. More judgement is purged, short-term numbers continue improving, more market rewards. Until the accumulated directional errors grow too large to ignore — but by then, the executives who made the purge decision have probably already walked away with their options. Klarna completed this cycle in two years. More companies are on the same path.
When the skeleton of a culture is hollowed out, it makes no sound. The earnings report will not tell you. The stock price will not tell you. The only thing that will tell you is the errors eighteen months later that should never have been made. But no one will attribute those errors to today's shuffle, because the causal chain is too long and the noise in between is too great.
This is the deepest wound of the cleaver — not the one inflicted on the person who was laid off, but the one inflicted on the organisation's capacity for judgement, perfectly sutured shut by a beautiful earnings report.
The Corner AI Cannot Illuminate
Pull Part III together.
Chapter 7 addressed the distortion of the measurement system — how education and certification use the wrong indicators to measure ability, causing people with genuine judgement to be overlooked. Chapter 8 addressed what AI exposes — when production costs reach zero, consciousness, architectural thinking, and design judgement become the only meaningful differentiators, things that had long been concealed. This chapter addresses why financialised corporate culture cannot be solved — even if AI breaks through every technical bottleneck, the costs of consensus, battles over discourse power, post-hoc attribution, and institutional misalignment of responsibility will remain exactly where they are.
Education cannot recognise true ability. AI exposes where true ability lies. And in corporations that have already been financialised, culture prevents capable people from exercising that ability.
Three layers of distortion stacked together, forming a closed loop.
One thing needs to be made clear: the primary cases dismantled in this chapter are from financialised companies in the tech industry — those that ignore their core business, focus on valuation, and make everything serve the numbers. But the dead knot of corporate culture takes a thousand forms; this is not the only one.
Traditional industries have their own variants. Industries whose entire value chain depends on government initiative — subsidies, policies, licences, relationships — have decision bottlenecks driven not by discourse-power battles but by approval hierarchies and political winds. The rhetoric is packaged for a different audience, but the logic of "structural problems translated into individual responsibility" is identical. People who see problems do not dare speak up, not because they fear losing discourse power, but because they fear offending those above. Post-hoc attribution does not pursue "who failed to persuade the team" but "who failed to read the leadership's intent." The form differs; the dead knot is similar.
This chapter cannot exhaust every variant. The constant across all variations is this thread: an organisation's decision-making mechanism will systematically suppress judgement, and AI accelerates building but cannot accelerate consensus, resulting in the exposure of structural waste that has always existed within the organisation. This thread is most nakedly visible in financialised tech companies, but it is not the tech industry's exclusive property.
Then, when education cannot recognise ability, AI exposes ability, and corporate culture traps capable people — after the three layers of distortion form a closed loop, one question remains unanswered: when the capacity for production itself approaches free, what do humans actually still need?
Education cannot recognise ability. AI exposes where ability lies. Corporate culture traps the able. Three layers of distortion stacked together, forming a closed loop. But one question remains unanswered: when the capacity for production itself approaches free, what do humans actually still need?