Skip to main content

Author: Nebula Walker Date: 28JUL2026 MYTHOGEN ENGINE (mythogenengine.com)

📌 Layoffs and record profits happening simultaneously. AI isn't the cause; it's the license. One exit, two sales: upstream to the hand pushing the button, downstream to the people under the button.

The Efficiency Trap · Sequel: You Thought You Won

Prologue: Xiaolin's Three Lessons​

Act One​

The day the senior engineer left, the office was unnervingly quiet.

Xiaolin remembered what the announcement said: "pursuing other career plans." No farewell party, no handover documents. A man who had been there for fourteen years cleared out his desk in a single afternoon.

The signs had been there all along—only no one had been willing to read them. Over six months, the senior's projects were reassigned one by one. His name quietly vanished from cross-departmental meeting invitations. When a production incident occurred—a decision that had never been in his hands—his name appeared in the responsibility column of the postmortem report. He appealed once. Then he never appealed again.

A month after the senior left, Xiaolin was promoted. His annual salary jumped from two million to three million. His manager clapped him on the shoulder and said: "The company believes in you. Young man, this is your era."

That evening, Xiaolin took his cohort out for dinner. He felt like he'd won. What he didn't know was that the person who'd held the position before him had earned ten million a year. The company had filled a ten-million hole with three million, netting seven million in savings on paper—and he was the name attached to that saving.

Act Two​

After taking over, Xiaolin discovered the system ran deeper than he could fathom.

Some architectural decisions he couldn't understand—why it was designed that way wasn't written down in any documentation. Some incident-handling records had only conclusions but no process—because the process had lived in the senior's head, and the senior was gone. He started working overtime, started losing sleep, started staring at his screen in the empty office late at night.

At the quarterly review, the manager's tone was still gentle: "Xiaolin, your potential isn't the problem—it's just that in the AI era, your thinking needs to keep up." After the meeting, the manager privately sent him a link: "Take this course. The instructor used to be at our company—he really understands our industry."

Xiaolin clicked the link. The course title glittered in gold, with an early-bird price of 4,999 yuan. In the instructor's photo, the man wore a well-tailored suit and smiled with composure—

It was the senior.

He could afford it. The extra million from the promotion made 4,999 yuan seem like nothing.

Xiaolin paid. The online course ran three and a half hours—prompt engineering, workflows, survival rules for the AI era. The content was shallower than he expected—so shallow he wondered if he'd missed some advanced modules. But he didn't dare say so. Because the comment section was full of familiar faces—his colleagues, contacts from the next department, even the manager who'd sent him the link—all had clicked "Like."

In the final chapter of the course, the senior looked into the camera and said: "In the AI era, the greatest danger isn't being replaced—it's refusing to change. Embrace change."

The audience—if an online course can have an audience—was filled with his former company's junior staff.

Act Three​

Three years later, the company announced it was being acquired.

The press release was beautifully written: three years of continuous workforce optimization, operating costs down 30%, profit margins at historic highs—an acquisition target with "first-class asset quality and technical team." The acquirer's CEO told reporters he was buying "the future."

Three months after closing, the new owners launched "organizational integration."

The day the list came out, Xiaolin stood in the corridor outside the conference room for a long time. His name was in the second batch. HR's wording was polite and standard: business overlap, structural adjustment, thank you for your contributions. Severance was at the statutory minimum—not a cent more.

He suddenly remembered many things. The afternoon the senior cleared out his desk. The manager's hand on his shoulder. "The company believes in you." The senior's composed smile in the 4,999-yuan course.

It turned out the senior hadn't been teaching him how to survive.

The senior had been demonstrating how he himself survived—with their tuition fees.

The company fired the senior, saving ten million; the senior launched a course, earning money from those who stayed; and the reason those who stayed were anxious enough to pay was precisely what the company's rounds of layoffs had taught them. Company, instructor, platform—all three got what they needed, running smoothly. The only one continuously paying was every Xiaolin who thought he'd won.

Even the raise that made him feel like a winner—part of it, passing through his hands, circled back into this chain. The extra money the company paid him sustained the courses that kept him too afraid to leave.

Outside the corridor window, it was getting late. Xiaolin took out his phone and scrolled through social media out of habit. The algorithm served him an ad:

"AI Career Pivot Bootcamp—early-bird discount, last three days. Don't get left behind by the times."

In the instructor's photo, the man wore a well-tailored suit and smiled with composure.

Xiaolin stared at that face for a long time, then pressed "Save."

On the chessboard, there are no allies or enemies—only pieces that haven't been exchanged yet.

The above is fiction. But the numbers that follow are not.

Chapter One: Take the Story Apart, Inside It's All Cash Flow​

The prologue is fiction. But if you tag every sum of money in the story and trace the direction it flows, you'll find this isn't a story at all—it's a cash flow statement.

The First Sum: Seven Million, and Its Multiplier​

The company fired the senior, saving ten million in annual salary; promoted Xiaolin, paying an extra one million. On paper, this position nets seven million in savings each year.

If the story ended here, it would be nothing more than routine cost control. The real key is this: that seven million doesn't exist at a value of seven million.

When enterprises are sold, valuations are typically calculated as a multiple of profit. Using the tech industry's common range of ten to fifteen times, every dollar of annual cost savings is magnified into ten to fifteen dollars of valuation at the moment of the transaction. The seven million saved from the senior's position is worth seventy to one hundred and five million at the negotiating table.

This is why layoff targets are always the highest-paid people—not because they contribute the least, but because their salaries are the biggest lever in the valuation formula. Firing one ten-million-a-year senior engineer contributes more to the transaction price than firing ten one-million-a-year juniors. And the systemic gaps, the loss of experience, the technical debt that won't detonate for three years? Those aren't in the formula. The formula only calculates up to closing day.

In other words: the senior wasn't replaced by AI. The senior was exchanged out by a valuation formula. AI was merely the exchange rate cited in the explanation.

The Second Sum: The One-Million Raise, and a Liability Booked as Profit​

Xiaolin's annual salary went from two million to three million. He gained an extra million, taking on a position nominally worth ten million.

But we can't simply say "Xiaolin delivered ten million worth of labor." The KPIs on the report were the same—projects ran, systems operated, targets were met—but his actual capabilities were not the same. The senior's ten-million salary had never bought just hours of work. It bought decades of accumulated judgment: why things were designed that way back then, how the last incident was resolved, which corners look like they can be cut but absolutely must not be. These things don't appear on the KPI dashboard, because dashboards only measure visible output—deep capability only surfaces the moment things go wrong.

So the seven-million "saving" actually consists of two parts. One part is the genuine extra labor Xiaolin contributes—overtime, shouldering pressure, burning himself out to fill the experience gap—this is extracted labor value. The other part is the capability gap that no amount of effort from Xiaolin can close—no one delivers this portion, yet it's booked as savings, turned into profit, turned into valuation.

In other words, half of the seven million on the company's books is squeezed out of Xiaolin; the other half is fundamentally a liability—a liability booked as profit. It won't appear on any quarterly earnings report. It will surface one day as an incident, a delay, a "how come nobody knows why this system was built this way." And on that day, the person standing at the scene, name written in the responsibility column of the postmortem, will be Xiaolin—just as the senior experienced in the prologue.

So Xiaolin's "promotion" is essentially two things happening simultaneously: he delivers labor below its value, subsidizing a deal he's not on the beneficiary list for; and he takes on an invisible liability for the company, waiting for it to detonate and serving as the on-scene responsible party.

And he's grateful for it. This is the most elegant part of the entire mechanism: the person being exchanged, at the moment of exchange, experiences victory.

There's one more link. The gap he can't fill doesn't sit quietly in the system—it reminds him every day: you can't handle this. That feeling has a name, anxiety; and anxiety happens to be the easiest emotion to monetize in the market. The capability gap Xiaolin can't close ultimately flows, in the form of 4,999 yuan, into the senior's pocket—and since the prompt engineering and workflows taught in the course can't fill a hole made of decades of judgment, the anxiety doesn't disappear, it only accumulates until the next checkout. The liability the company booked as profit has come full circle, becoming the senior's revenue.

This is the third sum of money.

The Third Sum: 4,999 Yuan, and Zero Customer Acquisition Cost​

Now look at the senior's course.

Anyone who's run a business knows that the most expensive part of selling courses isn't producing the content—it's customer acquisition. You have to spend money on ads, nurture traffic, build trust before a stranger will part with their cash. Marketing costs eating 30–50% of the price tag is the norm.

But the senior's customer acquisition cost approaches zero. Because his customers aren't brought in by advertising—they're delivered by the company.

Every round of layoffs the company conducts creates a batch of anxious potential students for him. Every time a manager privately forwards the course link, it's a precision referral. Every "Like" from a colleague in the comments section is a trust endorsement. The company manufactures fear; the senior sells the cure. And the cure can be priced far above the content's actual value—because what the students are buying was never those three and a half hours. It's the feeling of "I did something, so I'm temporarily safe."

The three parties complete an unsigned division of labor: the company saves millions in salary and gets a workforce that "self-funds continuing education and actively embraces AI"; the senior earns back the salary taken from him, one bill at a time, from his former juniors' pockets; the platform takes its cut and recommends the course to the next Xiaolin.

Nobody breaks the law. Nobody lies. Every party simply makes the most rational choice from their position. This is precisely what makes it terrifying—the structure doesn't need villains, only for everyone to be rational.

There's one more detail that only becomes visible when placed on the full ledger: the reason Xiaolin can afford this course is precisely because the company gave him that extra million.

A person earning two million a year, barely breaking even on living expenses, would hesitate at 4,999 yuan. A person who just got promoted to three million won't. The raise does two things simultaneously: it makes him believe he's on the rise, and it gives him disposable income—and what the anxiety market needs is precisely "anxious people who can afford to pay." Without that extra million, he's just an anxious person; with it, he becomes a customer.

And so the money completes a full circle: the company saves seven million from this position, allocates one million of it to Xiaolin, Xiaolin transfers part of it to the senior—and the senior is the very source of the company's seven-million saving. Money leaves the company, passes through Xiaolin's pocket, and returns to the person the company fired—while the company nets six million and gets a bonus: an employee who "self-funds continuing education and actively embraces AI."

The crucial point is: after completing this circle, Xiaolin's situation hasn't changed at all. He's still doing ten-million worth of work, earning three million, carrying the unrecoverable capability gap. The tuition he paid buys back none of his lost bargaining power. The raise, the learning, the professional development—at every step he did the right thing, and at every step he sank deeper into the position he was already in.

The Fourth Sum: The Invisible One—Public Funds​

The story doesn't mention it, but in reality there's a fourth stream of money.

The ecosystem surrounding these courses is often ringed by various lectures, meetups, and annual conferences. Nominally priced at 2,000 yuan, in practice you can attend for free by surrendering your personal information—because what the organizer really wants isn't your money, but your registration record. Your email, your attendance, your headcount—these become "service person-counts," appearing in the final reports for government subsidies.

Government subsidies act as an amplifier in this structure: they allow organizers to use public funds to invite industry speakers; the speakers' reputations build credibility; credibility attracts more headcounts; more headcounts secure the next round of subsidies. A business that was originally constrained by the scale of private capital has its ceiling removed once connected to public funds.

And so the full version of this cash flow is: taxpayers subsidize the events, events feed the traffic, traffic converts into course sales, and the existence of course revenue in turn proves that "demand for AI education is booming"—becoming the basis for applying for the next round of subsidies.

Closing the Ledger​

Now put all four sums on the same table and ask the simplest question: who is the net outflow party?

The company is net inflow: salary savings converted into valuation. The original shareholders and management are net inflow: transaction premiums pocketed. The senior is net inflow: salary recovered in a different form. The platform is net inflow: commissions and traffic. The government spent subsidies, but the money spent was taxpayers'.

The net outflow parties are only two kinds of people: the Xiaolins who paid tuition, contributed the labor price differential, and were ultimately severed at the statutory minimum; and the next wave of shareholders and employees who, after closing, inherit a hollowed-out organization.

The direction of cash flow is the direction of power. Throughout the prologue, Xiaolin couldn't understand his own situation—not because he wasn't smart enough, but because he was always looking upward from the outflow end. From that angle, every dollar flowing out looks like an opportunity.

Next, we shift the lens from the fictional Xiaolin to real numbers. In the first half of 2026: one hundred and fifty thousand Xiaolins.

Chapter Two: One Hundred and Fifty Thousand Xiaolins—When Layoffs and Record Earnings Appear in the Same Report​

Before diving into the numbers, we must honestly confront a methodological objection.

The prologue and Chapter One strung six links into a chain: companies fire expensive staff and hire cheap replacements; "natural attrition" masks forced departures; capability gaps are booked as profit; anxiety is monetized through courses; public funds amplify the entire market; and it all serves a transaction. If these six links are merely inferences, the entire chain is a slippery slope—the first link holding doesn't mean the second necessarily follows. No matter how smoothly the story reads, it cannot substitute for proof.

So what this article does from here on is take each link out individually and nail it down with verifiable public evidence: earnings reports, regulatory filings, named research studies. This chain isn't inferred—it's verified, link by link.

Start with the aggregate.

According to tracking data from tech recruitment platform TrueUp, by mid-June 2026, the global tech industry had seen 363 layoff events affecting nearly 150,000 workers—an average of 974 people per day, 44% faster than the same period last year. And this is only the first half.

Looking at the numbers alone, you might assume this is a recession. But the 2026 wave differs fundamentally from every previous round of layoffs: these companies aren't losing money. Many of them are delivering their best financial results in history.

A few examples, all from public earnings reports and mainstream media coverage:

Internet infrastructure company Cloudflare, in the same earnings call, announced cutting 1,100 workers (20% of its workforce) while posting record quarterly revenue of $639.8 million, up 34% year-over-year. CEO Matthew Prince specifically emphasized: this is not cost-cutting. The company's internal AI usage had grown sixfold in three months.

Enterprise software giant Salesforce cut approximately 5,000 customer service positions across two rounds. CEO Marc Benioff's exact words were that he "needs fewer headcount," while at the same time the company's AI customer service product Agentforce was already handling 1.5 million customer conversations. The positions cut were entry-level.

Oracle laid off approximately 21,000 people over the past twelve months and publicly attributed the cuts to AI investment, while forecasting that AI would bring further reductions. Amazon eliminated 30,000 corporate positions over several months. Meta, in the same quarter it reported substantial revenue growth, laid off 8,000 people while directing tens of billions of dollars in capital expenditure toward AI infrastructure. Cisco posted record quarterly revenue of $15.8 billion while cutting 4,000 positions.

Put these together and the pattern is almost brutally clear: record revenue, record profit, record capital expenditure—and layoffs. The money hasn't disappeared. According to multiple media tallies, the major tech giants' combined capital expenditure commitments toward AI infrastructure in 2026 total approximately $700 billion. The salaries saved through layoffs are immediately converted into budgets for chips, data centers, and model training.

The money also went somewhere else: shareholders' pockets. Official data from S&P Dow Jones Indices shows that S&P 500 companies' stock buybacks breached $1 trillion for the first time in the twelve months ending September 2025, setting a historic record; the pace of buybacks in 2026 continues at record levels, with full-year expectations for another all-time high. The buyback mechanism is worth explaining once: the company purchases its own shares and retires them, causing earnings per share (EPS) to rise even when total earnings remain completely unchanged—this is the most direct fuel for executive compensation and stock price. The cost savings from layoffs and the EPS boost from buybacks mutually reinforce each other on the same earnings report.

What Kind of Layoff Is Reasonable​

Before continuing, the strongest counterargument must be stated in full—otherwise nothing that follows will stand.

The counterargument goes like this, and it's quite powerful: when companies streamline their workforce, it's often genuine structural improvement, unrelated to AI or financial engineering.

Take the hardware industry as an example. A company that previously ran a "product proliferation" strategy—simultaneously maintaining thirty product lines, each with its own hardware team, drivers, testing, and supply chain contacts—had massive duplication across those lines. The same function written thirty times, the same problem solved thirty times, the same documentation maintained in thirty copies. When the company decides to consolidate into two shared platforms, keeping differentiation only at the outermost layer, the redundant work genuinely disappears. At that point, fewer people are truly needed—and not because "fewer people are doing the same work," but because "there's less work to do."

This kind of streamlining is correct. It doesn't just save costs—it actually makes the product better, because force previously scattered across thirty lines is finally concentrated on two. The software industry has equivalent versions: replacing fifteen homegrown tools with a single standard platform, or consolidating each team's independently maintained deployment scripts into unified infrastructure. These are genuine efficiency gains, and the accompanying headcount reduction is a reasonable outcome.

So the question was never "are layoffs always wrong." Layoffs can be right. The question is: how do you tell which kind?

There's a test—so simple it's almost crude, but very hard to fake: look at whether the workload actually decreased, and look at who left.

Genuine architectural consolidation has two inevitable characteristics. First, remaining employees' workload should stay flat or decrease—because the redundant parts have been eliminated. If, after consolidation, those who remain are working more overtime, carrying three people's load, handling at midnight what three teams used to share—then the redundancy hasn't been eliminated, only transferred to individuals. The work hasn't decreased; only the number of people bearing it has.

Second—and harder to fabricate: designing a platform that covers thirty scenarios requires more senior judgment, not less, than maintaining thirty independent product lines. To consolidate thirty lines into two, someone must know why each of those thirty lines was designed the way it was, which differences are essential, which are legacy baggage, and which one, if cut, will explode two years later. This knowledge exists only in the heads of people who've been there long enough. So a genuine architectural consolidation should retain precisely the most senior, most expensive people; what gets trimmed should be the redundant mid-level and junior work.

And so the test becomes very clear.

If a company announces structural streamlining, then retains senior architects, trims the redundant execution layer, the remaining staff's workload decreases, and two years later the system is genuinely more stable—that's real consolidation. This article's criticism has nothing to do with it.

If a company announces structural streamlining, and the first to disappear are the highest-paid, most senior people, the remaining staff's workload increases rather than decreases, and the new "platform" keeps breaking over the next two years because nobody knows why the original was designed that way—then what happened isn't consolidation. It's replacement. Streamlining is just the language it uses.

Chapter Three will show which answer this test yields against real data: under mandatory return-to-office and "natural attrition" policies, those who leave aren't the redundant execution layer but senior employees, high-skill workers, and senior managers—precisely the people architectural consolidation most needs to retain.

One sentence to summarize this section: When the work to be done actually decreases, that's efficiency. When the work to be done hasn't changed but the people doing it get cheaper, that's arbitrage. Both use the same vocabulary in press releases, both enter the same line item on the income statement, but three years later they leave completely different things in the system.

The Third Possibility: The Work Didn't Disappear—It Just Moved Off the Report​

The previous section distinguished two categories: genuine efficiency, and arbitrage. But in practice, the most common case is actually a third category—and it's the hardest to expose, because it leaves almost no trace on the income statement.

That is: converting in-house work to outsourcing.

The operation is simple. Work previously done by internal teams—driver development, testing and verification, operations, customer service, certain software modules—is handed over to external vendors, ODMs, offshore teams, or contract workers. Internal headcount drops significantly, and the drop looks great.

The key is that these two things go into different line items.

Full-time employees' compensation is booked under personnel expenses, and "employee headcount" itself is a public figure that investors watch closely—disclosed in annual reports, compared by media. Cutting 10% becomes an achievement to announce. Outsourcing costs are different: they go into cost of goods sold or operating expenses under professional services, outsourced services, or procurement—mixed in with a pile of other expenditures. There's no separate line item called "we moved this work outside."

So the same thing gets two narratives. The internal and external story is: we adopted AI and automation, achieving the same output with a leaner organization—workforce efficiency up 30%. What actually happened: the workload hasn't decreased at all. The people doing it switched from on-payroll to off-payroll; the cost moved from one line item to another. AI didn't do anything, but the credit goes to it—because announcing AI results at this moment makes the stock price rise; announcing "we outsourced our R&D" doesn't.

Can this be verified? Yes, and you don't need insider information.

Open the same annual report and plot the change in employee headcount alongside the change in outsourcing/professional services/procurement spending. If headcount drops sharply while outsourcing spending rises by a comparable amount, then the "efficiency from AI" needs recalculating—genuine efficiency gains should see both lines trending down, not one down and one up. This comparison is tedious to do, so few people bother; but it's all public data, available to anyone.

And the cost of outsourcing surfaces later than arbitrage, and is harder to reverse.

Replacing expensive people with cheaper ones at least keeps the capability gap inside the company—theoretically recoverable. Outsourcing moves knowledge entirely outside the walls: three years later, no one on payroll knows why that module was written the way it was. When the vendor raises prices, you have no bargaining power because switching costs are prohibitively high. When you try to bring it back in-house, you discover there's no one in the company who can take it over—you've even lost the ability to judge whether the vendor's quote is reasonable. What you saved was salary; what you sold was optionality.

This point will resurface in Chapter Five. Because a company that has outsourced its core capabilities and looks extremely lean on paper is a perfect acquisition target—until the buyer discovers what they bought isn't a team, but a stack of contracts.

Chapter One hypothesized that companies fill expensive positions with cheaper people. The evidence for this link comes from both ends of the labor market.

The exit end: labor analytics firm Revelio Labs' data shows that among white-collar workers who changed jobs in late 2025, 40% accepted pay cuts exceeding 10%—the highest proportion in at least a decade. Researchers call this "wage scarring": once you re-enter the market at a lower price, future compensation anchors to the depressed level long-term. Job platform ZipRecruiter's quarterly survey similarly found that over a quarter of new hires earned less than in their previous role, while those willing to negotiate dropped to 30%.

The entry end is more direct: after cutting a quarter of its workforce in 2022–2023, Meta was confirmed by media to be rehiring some of the same laid-off employees—at lower compensation. The company even set up an "alumni portal" for former employees to reapply. Same person, same skills, fired once, price reset once. And this playbook isn't even an AI-era invention: back in 2007, American electronics retailer Circuit City publicly announced it was firing 3,400 employees whose "pay exceeded market rates" and inviting them to reapply for their original positions at lower wages after ten weeks. Back then, doing this made headlines and triggered boycotts. Today, the same logic, packaged as "organizational transformation," is treated as good news for the stock price.

Something must be stated clearly here: this article does not deny the real productivity changes brought by AI. The issue isn't whether AI is useful, but that the word "AI" has changed its function in earnings-report language.

In the past, layoffs required reasons: economic recession, pandemic adjustments, business contraction. These reasons were negative—the company had to bear reputational pressure, and stock prices usually fell. Now, a single phrase—"AI transformation"—turns layoffs from bad news into good news. It proves the company is at the forefront; it proves management has vision. Stock prices don't fall—they rise.

AI isn't the cause of layoffs. AI is the license for layoffs.

There's also a distributional issue hidden beneath the aggregate numbers that deserves even more alarm: this wave of layoffs didn't land evenly. Entry-level positions bore the heaviest impact. Data from Stanford's Human-Centered Artificial Intelligence Institute (Stanford HAI) shows that employment among software developers under 26 has dropped nearly 20% since 2024. Cryptocurrency exchange Coinbase, during its restructuring, directly eliminated "pure management" roles, reorganizing into small teams led by senior engineers directly commanding AI agents—a structure with no room for newcomers who need guidance, who need time to grow.

Place this back into the prologue's logic and you see a longer time bomb: Xiaolin at least had the chance to sit in that seat. The next generation of Xiaolins won't even get a ticket through the door. If nobody hires junior engineers today, there will be no senior engineers in ten years—but as Chapter One noted, ten years from now isn't in the valuation formula. The formula only calculates up to closing day.

Chapter Three: "Natural Attrition"—The Cleanest Word in the Earnings Report​

If you read these companies' earnings reports closely, you'll notice a curious phenomenon: many companies never once use the word "layoff."

In its place is a cleaner vocabulary: "employee reorganization," "hiring freeze," "organizational optimization," and the most elegant one—"natural attrition."

"Natural attrition" sounds like a weather phenomenon: employees evaporating like moisture, the company simply choosing not to replenish. No layoff announcements, no severance payouts, no media headlines—it doesn't even count as a layoff statistically. Year after year, headcount quietly drops by 10%.

But anyone who has worked at a large enterprise knows that beneath the word "natural," there is often a great deal of engineering that is anything but.

This link is the easiest to dismiss as conspiracy theory, which is precisely why it has the most solid evidence—because the ones who admit it are the executives themselves.

HR software company BambooHR, in a survey of over 1,500 U.S. managers, found: one-quarter of senior executives admitted that when implementing return-to-office (RTO) policies, they hoped it would trigger "voluntary resignations"; one-fifth of HR professionals admitted the company's in-office requirements were designed from the start to make people quit; and nearly 40% of managers believed their company later conducted formal layoffs because not enough people had resigned under RTO. The report's conclusion used a blunt phrase: RTO is "layoffs in disguise." This isn't a media accusation—it's self-reported by the parties involved. Even the U.S. Federal Reserve's Beige Book has recorded employers using return-to-office mandates to "encourage attrition" as a deliberate workforce reduction strategy. Another 2025 survey of over 1,100 business leaders found that more than half of U.S. companies admitted to using "quiet firing" tactics—stalling raises, increasing attendance requirements, cutting benefits—to push employees to leave on their own.

And who does this method drive out? University of Pittsburgh scholar Mark Ma's team, after analyzing career data from millions of professionals, found: under mandatory in-office policies, those who leave aren't low performers but the top talent—the attrition rate among senior employees, high-skill workers, and senior managers rises to 18–19%. In other words, the sieve of "encouraged attrition" filters out precisely the most expensive people with the most market optionality. This aligns perfectly with Chapter One's valuation logic: those who were going to leave were always the ones with the largest salary leverage.

At the individual level, the operations were already described in the prologue's Act One: projects reassigned one by one, names quietly removed from meeting lists, responsibility for others' decisions placed on your shoulders. These tactics share one common feature—each incident, viewed in isolation, has a reasonable explanation. Together, they form an unmistakable message: this place no longer needs you. The target usually submits their resignation before their dignity runs out, saving the company severance costs. On the statistics sheet, it becomes one more entry under "natural attrition." Everything legal, everything clean.

And those executing these operations are often not HR—they're the next generation looking to move up, with upper management's tacit approval or even guidance, isolating and hollowing out the predecessors to claim their positions.

Here's a question worth every person in this situation calmly calculating: how long does this path upward actually pay?

The answer is buried in the timeline. Refer back to Chapter Two's data: these companies' workforce reductions aren't one-time events—they're rolling compressions, one round per year, continuously ongoing. This means that the person who cooperated in pushing out the predecessor this year is already inside next year's compression base. He has proven two things to the company: first, his price is far cheaper than the predecessor's; second, he's willing to cooperate in doing this. The first makes him a target for the next round of valuation optimization; the second removes any psychological burden for the company—because the same tactics can be executed by someone even younger and cheaper.

He thought he'd boarded a helicopter. In reality, he's sitting in a chair with a timer, and the timer isn't his to set.

Old Zhou from the Next Team Over​

The following is fiction.

Xiaolin's neighboring team had a guy everyone called Old Zhou. Old Zhou wasn't any kind of star employee—he was the kind you don't notice, but the system breaks without him. The modules he was responsible for, the parts the documentation couldn't explain clearly, all depended on his memory. When things went wrong at 2 AM, he was the one who got called. He'd been at the company twelve years; his salary ranked near the top of his department.

The changes happened slowly.

First, the projects. A core module Old Zhou had owned for three years was transferred to another team during an "architectural optimization." The reason was reasonable: cross-departmental integration. Old Zhou was assigned to a new project with vague scope and delivery criteria no one could articulate.

Then, the meetings. Old Zhou noticed he was no longer being invited to certain regular meetings—not notified "you don't need to attend," but the meeting link simply stopped appearing on his calendar one day. He asked once. The answer: "Oh, that was a system glitch. I'll add you back." The next week, the link disappeared again. He didn't ask again.

Then, the performance reviews. In his quarterly evaluation, Old Zhou encountered a phrase he'd never seen before: "collaboration needs improvement." He went to his manager. The manager's tone was gentle: "This isn't just my opinion—it's multi-source feedback." Old Zhou asked whose feedback. The manager said: "I can't disclose that, but maybe think about whether there's been any friction in your recent cross-departmental work."

Old Zhou thought about it for a long time. He couldn't think of anything. But he began doubting whether he'd genuinely missed something.

Over the next few months, the pattern repeated: his name disappeared from one review list after another—not crossed out, but simply "forgotten" when new versions of the lists were compiled. Once, a problem arose from a decision he'd never been part of, and his name appeared in the postmortem report. He checked the meeting minutes and found he'd never been invited to that meeting. He brought the evidence to his manager. The manager looked at it and said: "I understand how you feel, but right now what matters is solving the problem, not assigning blame."

Old Zhou noticed something: no single person was doing this. When the meeting link disappeared, the manager didn't know—or at least, appeared not to know. When the list "forgot" to include him, the person responsible apologized sincerely. The "multi-source feedback" on performance came from different departments; each person, asked individually, said "I was just being honest." No one was lying. No one admitted to coordinating. But everything, taken together, pointed in the same direction.

He started losing sleep. He started feeling dread on Sunday evenings. He started compulsively checking his email, making sure he hadn't missed something. He went to a doctor, who said it was anxiety disorder and prescribed medication. He didn't tell any colleagues, because he wasn't sure he was overthinking—after all, each incident taken alone had a reasonable explanation.

Three months later, Old Zhou submitted his resignation.

The announcement said: "pursuing other career plans." The same phrasing as the senior's announcement in the prologue.

When Xiaolin heard the news in the break room, he said: "Old Zhou was a good guy. Too bad." Then he went back to making his coffee.

The above is fiction.

But if you've ever been at a company that was "optimizing its workforce structure," you might find this passage carries an uncomfortable familiarity. If you feel uncomfortable—take that feeling seriously.

"Natural attrition" is the cleanest word in the earnings report. And beneath clean words, sometimes what's buried isn't just a person's position—it's a person's integrity. Their trust in their own judgment, their belief in the basic fairness of the world, their ability to live as a normal person. When these things are stripped away layer by layer, and each layer's removal has a "reasonable explanation," the victim ends up doubting not the system, but themselves.

There's also a numerical tell—one that, once seen, can't be unseen.

"Natural attrition," as the name implies, should be natural—employees leave for personal reasons, family, health, career changes, retirement, at a roughly stable rate. A thousand-person company in a normal year has a natural attrition rate of roughly 8–12% (varying by industry). This number doesn't change because of the company's financial targets, just as rain doesn't stop because you need sunshine.

But open the data of these companies that are "optimizing workforce structure" and you'll find something impossible: the rate of "natural attrition" spikes dramatically precisely in the years the company needs to cut headcount. What used to be 10% per year suddenly becomes 20%, 30%. And the timing of the spike aligns precisely with the earnings report's needs—not too early, not too late, falling exactly in the quarters when cost targets need to be met.

Natural phenomena don't align with your budget cycle. Rain doesn't stop because you have a wedding tomorrow. Employees don't collectively decide to leave precisely because the company needs a prettier profit margin next quarter.

If the rate of "natural attrition" accelerates precisely when the company needs it to accelerate, then it's not natural. And if it's not natural, then what the earnings report records with the phrase "natural attrition" is not a fact—it's an inaccurate description packaged in accurate terminology. A clean word from a dirty source.

No one will be held accountable for this contradiction. Because "rising natural attrition" has countless explanations—a good market, competitors poaching, generational attitude shifts—each one reasonable, each one sufficient to deflect any inquiry. But you only need to ask one question: among all these explanations, which one accounts for why the attrition peak falls precisely in the quarter the company needed layoffs but didn't want to call them layoffs?

And a final sentence, for everyone who's read this far:

How far is this fictional story from the reality you've witnessed? If the answer is "not far"—then what stands behind that clean number in the earnings report? And in the quarter that number spiked, how many Old Zhous were there?

Chapter Four: The Cure Market—When Anxiety Becomes the Best-Selling Product​

Where did the people exchanged out of the system go?

Some changed careers, some went silent, and some—started selling courses.

This isn't hard to understand. A laid-off senior engineer needs to maintain income, and the market happens to have the most precisely targeted buyers: people still employed, having witnessed round after round of layoffs, terrified they'll be next. Fear is manufactured by companies for free; the instructor only needs to sell the cure. And so a complete food chain takes shape: companies use AI as the excuse to fire people; the fired come out and teach AI; those still employed pay out of pocket to learn—workers fund the training of their own replacements, and the company doesn't spend a cent on training.

Every segment of this food chain is supported by named survey data.

The fear end: HR consultancy Mercer's survey of 12,000 workers globally shows that in 2026, 40% of employees fear being replaced by AI—up from 28% two years ago. KPMG's survey of over 2,100 U.S. employees is even higher: 52% worry their job will eventually be taken over by AI, nearly double from a year ago. Fear isn't this article's rhetoric—it's a collective state that has been repeatedly measured.

The self-funding end: HR group Randstad Digital's late 2025 survey of 27,000 digital workers and more than 1,200 employers found that roughly half of tech workers are paying out of pocket for AI training because their employers won't prepare them. Online education platform edX's survey found that three-quarters of tech workers believe they must complete skill upgrades within six months to keep their jobs; over half planned to self-fund more than $1,000 on professional development that year.

And the most telling figure is the reverse data from the enterprise end: in the same period, employer spending on enterprise AI tools totaled approximately $32 billion, while investment in employee training was under $4 billion—an eight-to-one ratio. Research firm Gartner found that corporate training budgets were cut by an average of 18% in the second half of 2025, while AI tool spending rose 23%; LinkedIn's workplace learning report shows the percentage of companies offering formal AI training is dropping rather than rising. Money for buying machines is being doubled down; money for teaching people is being withdrawn—the gap is left for employees to fill themselves, and employees dutifully filled it. This is the complete evidence chain for "workers self-funding the training of their own replacements": companies manufacture fear, withdraw training, employees pay up—three datasets, one closed loop.

But before lumping all course sellers into one category, a line must be drawn. Because this market is actually a spectrum.

At one end of the spectrum are courses with real substance. I've personally paid 1,000 yuan and taken an online course teaching small businesses to build AI customer service chatbots: three hours of content, three months of replay access, examples that were genuinely usable, and the instructor even considered students' budget constraints, teaching how to use low-cost API plans. It wasn't perfect—the discussion of scalability and long-term costs lacked depth—but after finishing, you had a tool in your hands that actually worked.

At the other end of the spectrum is 5,000 yuan for three and a half hours, teaching you to use prompts to generate AI virtual avatars, bundled with a "prompt toolkit." After finishing, what's in your hands? A batch of AI-generated photos and a feeling of "I did something."

The problem is: from the marketing page, you can't tell the two ends apart. Glittering visuals, "Bootcamp" branding, countdown timers for early-bird pricing, the instructor's titles and experience—both ends use identical language. The instructor's credentials are even real: prestigious university master's degrees, big-tech engineering backgrounds, industry certifications. The authenticity of credentials here isn't proof of capability—it's sales packaging. Its function is to get you to click "Confirm" on the payment page.

And this market can't self-correct, because the reason is buried in the buyer's psychology: people who buy empty courses almost never say so publicly. Saying so means admitting they were harvested. Sunk cost makes victims choose silence, or even become advocates—"the course was still inspiring." So negative reviews of poor courses are perpetually scarce, marketing pages are perpetually glowing, and the next Xiaolin always sees nothing but prosperity.

A Conversation​

I have a real record spanning nine months, with a person I'll call A-Bao. Thirty years old, graduated from a technology university, no relevant degree, no industry experience, wanted to enter the AI field. Here is what happened between us.

This is how we met. In the period before generative AI matured, I'd been answering Python questions online intermittently for free—no agenda, simply because someone asked and I happened to know the answer. A-Bao added me this way; he wanted a learning roadmap.

My first piece of advice wasn't any course—it was a sentence: first figure out what you want to do.

The judgment behind this sentence was: AI and software are an extremely broad spectrum—data engineering, model training, application development, systems integration, product design, automation workflows—each path requires different capabilities, suits different personalities, and has different ceilings. And he didn't have a relevant degree, hadn't been on any of these paths. In this state, the worst thing to do is pay for a course first and let that course decide his direction—because courses sell what they have, not what he needs.

So I told him: find the direction first, then find the method. Direction requires experimentation, and the cost of experimentation can approach zero.

In September 2024, he asked me how to break in. I spent an entire message explaining: don't make AI development the core—start from applications, first understand your own business pain points, use the simplest AI to solve the most concrete problems. I sent him three links: Andrew Ng's DeepLearning.ai (free courses, industry gold standard), W3Schools (free technical fundamentals), and the open-source model library on Hugging Face. Each one free, each one an industry-recognized top-tier resource.

Five months later, in February 2025, A-Bao came back. His update: he'd earned a Google certification and was working toward a generative AI competency certification from an institution. He asked me if I was looking at any courses.

I told him: fifty bucks rents a cloud GPU, twenty-six minutes is enough to fine-tune your own model; Hugging Face has massive open-source models ready to use; hard technical fundamentals are entirely available in free resources.

His reply was: Got it. Then he asked if I'd gotten any certifications.

I stared at that sentence for a while.

What happened in that moment says more about the state of this market than anything else in the entire conversation: while I was teaching him, he was using a piece of paper I don't have to judge whether I was qualified to teach.

This isn't rudeness. He might even have been concerned—"you should go get one too." The real problem is that this was the only ruler he had left. Technical judgment, practical pathways, understanding of the tool ecosystem—he couldn't evaluate these things, because evaluating them requires precisely the capability he doesn't yet have. Certifications he can evaluate: have, or don't have. When a person has only one ruler, they'll use that ruler to measure everything—including things the ruler can't measure.

And so a strange result emerges: a person who spent months giving him top-tier industry resources for free is, in his cognitive framework, "not ready yet"; while a general-purpose, beginner-level certification is proof of "being ready."

I don't blame A-Bao. And to be clear: his choice, in his situation, is rational.

A thirty-year-old with no experience, no network—what he lacks most isn't knowledge. The three links I sent him contained everything he needed. What he lacks most is a ticket: a signal that an HR person recognizes when opening a resume. The certification gave him this signal. After passing, he went from "has nothing" to "certified"—starting salary could go from two million to two-fifty or even three million, and companies would be willing to give him interviews. His value genuinely increased because of that piece of paper. The system, at the moment of entry, is positive for him.

But the same piece of paper, placed at the other end of the spectrum, does the opposite.

An engineer with fifteen years of experience, if they also take the same certification, gets placed by HR's search system in the same field as A-Bao—same checkbox, same filter result. Fifteen years of judgment is invisible in that field, because the field only holds "have" or "don't have." So their bargaining power is compressed to the same level as all certification holders. Don't take it, and they can't even enter the field. Take it, and their value is flattened.

The same certification is elevation for those without experience, and compression for those with it. Both forces act simultaneously, with the effect of pressing a market that previously had price differentiation into a commodity market—and commodity prices always trend downward, because certification is designed for mass production: pass rates above 70%, eligibility is "everyone," and the goal isn't selection but supply. The more supply, the less scarcity each piece of paper holds, and the more leverage employers have to push prices down.

So what A-Bao gained is real. But at the same time he gained it, others are losing—and the way they lose is precisely A-Bao's existence itself: he is a labeled, cheaper alternative option. He doesn't know his role in this chain, just as Xiaolin in the prologue didn't know he was sitting in a ten-million-dollar chair.

Sweet and bitter dissolve in the same cup, inseparable. This structure doesn't need villains—it only needs everyone to make the most rational choice. And the cumulative effect of rational choices is that pricing power for the entire market transfers from experienced practitioners to an exam with a 70% pass rate.

And those links I gave A-Bao—free, top-tier, sufficient to take him from zero to independent work—in his world, are equivalent to nonexistent. Not because he can't see them, but because they don't have a price tag. Things without price tags, in a market that classifies everything by labels, cannot be recognized as valuable.

The fire extinguisher got a price tag. And the water that actually puts out fires, because it's free, doesn't count.

Finally, connecting back to the fourth sum in Chapter One: the peripheral ecosystem of this cure market—communities whose activity is padded with insiders' accounts, instructor networks that cross-endorse each other, meetups where personal data is the price of admission—is being amplified by public funds. This link also doesn't need inference: the subsidy mechanism is a matter of public policy. Taking Taiwan as an example, the Ministry of Labor's Industry Talent Investment Program subsidizes 80% of course fees for employed workers, with full subsidies for specific demographics; various ministries promote AI-related lectures, workshops, and industry events using "service person-counts" as outcome metrics—this is precisely the institutional root of why registration records and headcounts have value. The government subsidy's intent is to promote industry learning; its actual effect is to remove the scale ceiling from this anxiety-monetization machine. Taxpayers pay; the machine accelerates.

The Second Sale: Selling the Narrative Back to Decision-Makers​

Up to this point, everything this market sells is cures for employees. But it has another floor, priced higher, harder to verify delivery, and the buyer sits on the other side of the conference table.

I recently read a widely circulated industry analysis. The argument goes: AI's first act is the model war, the second act is the agent war, and the third act will be the deployment war. As models commoditize, value will shift from "who has the strongest model" down to "who can make compute cheap and small enough to fit in devices." And this direction happens to settle into the sweet spot of a certain region's strengths.

In fairness, the article's broad direction is insightful. The shift of value from the model layer to the deployment layer—I agree with that judgment. But take it apart for inspection and a structure emerges: the first four steps of reasoning are solid; the last two—"compute descends" to "the local industry therefore benefits"—are skipped over. An entire section of argument is missing: mastering manufacturing doesn't mean mastering the platform; being able to build something doesn't mean capturing value. The companies that actually made money always had moats that were more than hardware—hardware plus developer ecosystem, plus software stack, plus distribution channels. And the article barely touches on the most critical layer—above the model, below the hardware—the system layer responsible for orchestration, memory, tools, and permissions.

So the question becomes: why does an article with gaps in its argument circulate so widely?

Because its structure is a perfect fit for another purpose.

"First act, second act, third act"—that's the skeleton of a slide deck. "Model commoditization" is a single chart. "Value descent" is a downward arrow. "Local industry's opportunity" is the closing slide that makes everyone nod. The most likely product to successfully emerge from this article isn't any deployed AI system or edge device. It's a series of courses: "AI's Third Act: Industry Opportunities in the Deployment Era," "The AI Agent Era: How Enterprises Transform," "Enterprise AI Strategy: From Proof of Concept to Scale." Then consulting services, corporate training, and government subsidy project proposals.

This is the cure market's upgrade. Previously what was sold was "learn to use AI," priced at 4,999, with buyers being employees afraid of layoffs. Now what's sold is "understand the next decade of the AI industry," priced an order of magnitude higher, with buyers being senior executives afraid of making wrong decisions. The latter is more sophisticated and easier to sell—because its deliverable is "cognition," and cognition cannot be audited. An employee who finishes a course can at least check whether they can build the tool; an executive who attends a trend lecture—how do you prove they "understood" or didn't?

And there's a closed loop here worth pausing to see clearly.

Chapter Two explained that the "AI transformation" narrative functions as turning layoffs from bad news into good news—it's the license for layoffs. Now notice where this license comes from: trend analyses, industry forums, consultant briefings, executive training courses. That is, the same group of people first sell the story "AI is reshaping everything, you must transform" to decision-makers—decision-makers take this story and execute layoffs; then sell the story "in the AI era you must upgrade your skills" to the laid-off and the afraid-of-being-laid-off.

One layoff, two sales. Upstream sells to the hand that pushes the button; downstream sells to the people under the button. And both sides are buying two sides of the same story.

I don't believe the people doing this are consciously designing this closed loop. Trend analysts genuinely believe their judgments; instructors genuinely want to help; executives genuinely believe they're making the right transformation decisions. But sincerity doesn't change structure: when a narrative is simultaneously the rationale for a decision and the content of a product, the storyteller is simultaneously the consultant and beneficiary of the transaction—and no rule requires them to disclose this.

So the place where that article is most likely to be successfully deployed isn't any production line, any edge device. It's an executive's slide deck.

This also explains why, in this market, insightful articles don't necessarily outperform, and rigorously argued analyses don't necessarily get seen. Because what the market is actually purchasing has never been "accuracy"—it's "things you can present." Accurate analysis is often full of caveats, can't be wrapped up in a single sentence, and can't end with an empowering conclusion. Things you can present must have a three-act structure, a downward arrow, and a conclusion that makes everyone nod.

In a market where narrative itself is the product, the most efficient product has never been AI—it's the story about AI.

Chapter Five: Endgame—All of This, for One Transaction​

Now answer the most fundamental question of this entire article: why do companies do this?

Firing the most experienced people, letting hidden systemic gaps accumulate, draining team morale—no management team that genuinely intends to run a business long-term would systematically do these things. Unless long-term operation was never the goal.

The goal is a transaction.

And to put it more completely: the goal is to exit. And exit has only two paths.

The first is cashing out. The boss is getting older; the second generation doesn't want to take over, or takes over but can't sustain it; the core business's market is shrinking, returns thinning year after year; continued investment has no visible payback horizon. The most rational choice at this point isn't revitalization—it's dressing the company up and selling it. Every action to compress costs and beautify profit margins serves the same transaction; the valuation formula was already calculated earlier.

The second is pivoting. Not selling the company, but treating the core business as an ATM—no more investment, no more talent development, no more upgrading equipment and architecture. Cash squeezed from the core business is redirected to something new: real estate, investments, distribution, a new venture in a different sector. The old business isn't being managed; it's being harvested. Its function shifts from "creating value" to "generating cash flow for transfer."

These two paths look exactly the same from the outside: headcount in continuous decline, profit margins rising, investment shrinking, senior people leaving batch by batch, those remaining doing the work of three. From the inside they look the same too—nobody announces "we've decided to stop playing." Employees just feel resources getting tighter, decisions getting more short-term, questions about "what's the three-year plan" going unanswered.

And Hong Kong and Taiwan are no strangers to either path.

Over the past several decades, we've watched the same play countless times: factories moving north, production outsourced, the core hollowed out, assets redirected to real estate or financial investments. Legacy manufacturing brands still standing, but behind the signboard the business has changed. We've also watched large numbers of family businesses choose to sell because the second generation won't take over, or maintain an uninvested shell until it can be sold. The common thread in all these stories: the real decision was made years ago, and employees are always the last to know.

Every time, that decision needs an external narrative. And each era provides a different one: in the 1990s it was "westward migration, cost restructuring"; after the financial crisis it was "weak global demand, going lean"; later it was "digital transformation"; then "the new normal after the pandemic."

This time, the narrative is AI.

And the reason AI works better than the previous narratives lies in their nature: cost pressure, weak demand, global recession—these are all bad news. The company has to bear public scrutiny, and stock prices typically fall. Only this one is good news. Saying you're transforming makes the stock price rise. A company that has decided to exit was going to fire people anyway, going to stop investing anyway, going to let the most senior leave first anyway. AI didn't create this script, it's simply the first reason that makes executing this script look like progress.

This also explains something otherwise inexplicable: why companies that are laying off are simultaneously making record profits. A company that has genuinely become more efficient through AI should reinvest the savings—hire more people to do more things, expand the battlefield. A company that is exiting takes the savings and buys back stock, raises dividends, beautifies next quarter's EPS. In Chapter Two's data, which of these is actually happening—readers can judge for themselves.

Extend the timeline to five to ten years and the logic of the entire operation becomes immediately clear: continuously compress labor costs, profit margins rising year over year; the "AI transformation" narrative makes the cuts look like progress; stock buybacks and dividend increases push up the share price; every metric on the earnings report serves the same purpose—dressing the company up as a perfect acquisition target. Chapter One already calculated the formula: every dollar of annual cost saved is worth ten to fifteen dollars at the negotiating table. Three years of "optimization," exchanged for a one-time premium cash-out. As for the technical debt and capability gaps that won't detonate for three years? After closing, that's the buyer's problem.

Let's first nail down this link's premise: does that "liability booked as profit" actually explode? The answer from management research is quite consistent. A Harvard Business School research summary indicates that the short-term cost savings from layoffs are often offset by knowledge loss, declining employee engagement, rising voluntary turnover, and diminished innovation, ultimately harming long-term profitability. A 2022 study published in the Academy of Management Journal found that companies experiencing layoffs saw patent applications drop 30% in the following year. And in the AI-layoff era, this pattern repeats at faster speed: a 2026 compilation of multiple enterprise surveys shows that among companies conducting large-scale layoffs citing AI, approximately 55% subsequently admitted regret—quality deteriorated, institutional knowledge proved impossible to rebuild, and rehiring costs exceeded the original savings. The liability does explode—that's a documented pattern. The detonation timing is simply designed to fall after closing.

But the story has a second half, because the buyer isn't stupid either.

The party paying top dollar for the shell never intended to run the business from the start. What they're buying is the company's IP, client list, historical data, brand image—and most importantly, a ready-made listed-company status. After closing, the second round of stripping begins: valuable assets are gradually transferred—client contracts moved to affiliated companies, IP licensed to insiders, core teams "naturally attriting." Simultaneously, the shell is used as a platform for fundraising and share placements, or for reverse-injecting the buyer's own business to achieve a backdoor listing. Nominally it's "business restructuring" and "strategic transformation"; in reality, it's emptying whatever value remains in the shell.

"The buyer isn't buying a business—they're buying assets to extract"—this link sounds the most like conspiracy theory, so let's close with a case that has a complete public record. In late 2023, chip giant Broadcom completed its $61 billion acquisition of virtualization software company VMware. The subsequent operations are all documented in earnings reports and mainstream tech media: headcount was cut by over 19,000—more than half of the pre-acquisition workforce; perpetual licenses were terminated, forcing customers to switch to subscription models; price increases on the customer end ranged from 1.5x to 3x at the low end, with cloud service providers reporting complaints of 8x to 15x increases—U.S. telecom giant AT&T even sued, alleging quoted increases exceeded 10x (later settled). The result? Broadcom's infrastructure software division's operating profit margin rose from VMware's pre-acquisition self-operated margin of 10–20% to 77%. And the cost is also publicly forecast: Gartner estimates VMware's market share will drop from 70% in 2024 to 40% by 2029, and customer surveys show the vast majority of users are actively using, planning, or considering alternatives. Translated into this article's language: the buyer didn't run the business; the buyer is liquidating the business—converting decades of accumulated customer lock-in and technical moats into profit margin at maximum speed before they depreciate. This isn't a failed acquisition. By the buyer's objectives, it's an extraordinarily successful one.

And so the complete harvest comes in three rounds: Round One—original management fires staff and dresses up the books, sells and cashes out. Round Two—the buyer strips assets, extracts residual value, and cashes out. Round Three—there is no Round Three harvester; only Round Three bearers: the employees left in the shell, and the retail shareholders and customers holding the last hand.

If this script looks familiar, it's not a coincidence. In my Capital Game Players series, I fully deconstructed this standard operating procedure of capital using real cases from the tech and gaming industries: personnel replacement, acquisition dilution, targeted layoffs, AI narrative packaging—four steps, replaying across different companies and different industries, with even the sequence rarely changing. That series documented how this script swallows company after company; this article documents what the same script looks like when it lands on a specific person—the view from the afternoon Xiaolin stood in the corridor.

Before the Final Chapter: AI Is Not Fake​

If, having read this far, your takeaway is "so he's saying AI is a scam"—please stop here, because that is not what this article means.

AI's capabilities are real. It genuinely enables one person to produce what previously required three; it genuinely compresses the time for writing code, doing analysis, and processing documents to a fraction of what it once took; it is genuinely changing how every industry works. The analysis process of this very article used AI—if I didn't believe in its real productivity, I wouldn't use it every day.

And everyone should learn to use it. This point comes without any reservations. AI is the most powerful personal productivity tool of this era. Not learning it means giving up an enormous lever. This isn't a course advertisement's sales pitch—it's a fact.

So where's the problem?

The problem isn't AI. It's misalignment in four directions.

Companies are using it wrong. Genuine efficiency gains should mean using AI to help the people who remain do better, do more, create new things—then reinvesting the time saved into next-generation products and capabilities. But what many companies are doing is using AI's name to fire people, distributing the savings to shareholders, and declaring it transformation. Chapter Two's data already shows: the $700 billion in capital expenditure buys GPUs and data centers, not employees' futures.

Government subsidies are being used wrong. The purpose of subsidies is to lower learning barriers and address market failure. But when the subsidy mechanism uses "service person-counts" as its outcome metric, what it rewards is volume, not quality. An exam with a 70% pass rate and "everyone" as the eligibility requirement doesn't produce talent—it produces labels. And labels' effect is to flatten the entire market's pricing.

The course industry is using it wrong. AI is genuinely worth teaching, but the market's best-selling courses often sell not AI's capability, but the fear of being left behind. Fear-driven purchases don't care what the course teaches—they only care about the reassurance that comes from the act of "I bought it." So poor courses and excellent courses use identical marketing language, and the market cannot self-correct.

Media is using it wrong. AI is a genuinely good story, but good stories are too useful. It's being used to package layoffs, package transformation, package exit decisions—even being used to sell twice simultaneously, upstream to the decision-makers pushing the button, downstream to the people under the button. Media isn't lying—it's simply repeating a narrative that is simultaneously fact and commodity, without distinguishing between the two.

So what this article has never criticized is AI. What this article criticizes is: a genuine technological advance is being systematically appropriated as cover for a set of decisions that have nothing to do with technology. And the most irrefutable aspect is—the cover works precisely because the core it wraps around is real.

If AI were fake, all of this would actually be simple—just expose it. Precisely because AI is real, you need to look more carefully to distinguish: is the thing you're being pushed to do learning a real tool, or cooperating with someone else's exit that has nothing to do with you?

The answer is sometimes the former. Sometimes the latter. Sometimes both at once. And the method for distinguishing them—the previous five chapters have already provided it.

Final Chapter: The Biggest Blind Spot​

Having written this far, one question must be honestly confronted: has nobody truly seen any of this?

No. Many have seen it. Those inside the system feel it more clearly than any observer.

The real blind spot isn't in failing to see—it's in the conclusion that follows seeing. Most people's conclusion is: the rules are set by others, I can't change them, so the only rational choice is to join—sign up for that course, enter that circle, learn that set of talking points, and when necessary, cooperate in pushing out my own predecessor.

And that conclusion is the most important component of the entire machine.

It doesn't need you to truly believe. It only needs you to feel there's no alternative. When enough people join, those who don't are automatically categorized: behind the times, bitter, sore losers complaining about unfair rules. Critical voices don't need to be rebutted—they only need to be categorized. Once categorized, the voice is neutralized. And so the machine acquires its best defense: its victims are simultaneously its defenders.

This article doesn't offer a cure. The market has enough people selling cures, and you already know how cures are priced.

It only wants—the next time you see "record revenue" and "organizational optimization" in the same earnings report, the next time you receive a course link privately forwarded by your manager, the next time you feel like you've finally been seen and promoted—to leave you with an equation you can verify yourself:

Where does the money come from, and where does it go? Who is the net inflow party, and who is the net outflow party?

Then ask the question Xiaolin figured out standing in the corridor—but ask it before you reach the corridor:

Which link of the chain are you on right now—and which link do you think you're on?

This article is the sequel to "The Efficiency Trap." For a complete case analysis of the capital asset-stripping playbook, see the Capital Game Players series.

The companion piece, "The Reader in the Anomaly Report—When Truth and Garbage Enter Through the Same Channel," records the other half of the story: what happens to these words on the platform when you try to write them down. The two pieces together form the complete picture.

📚 The Efficiency Trap and Cognitive Erosion Quintet

  1. Cognition and Judgment—The Last Thing AI Cannot Replace
  2. How Environments Make People Foolish—Cognition and Judgment (Postscript)
  3. The Efficiency Trap: When Everyone Is Learning to Compress, Who Is Creating New Demand?
  4. The Standard Operating Procedure for Killing Innovation
  5. The Efficiency Trap · Sequel: You Thought You Won