All eight people below are invented. Every single thing that happens to them can be traced back to something real.
1. Wang Rui, 32, former head of operations
Wang Rui’s company was doing well. Gross merchandise volume grew thirty percent last year, and the first quarter this year came in profitable. They even replaced the coffee machine in the break room.
Then on a Friday afternoon in April, HR pulled him into a small meeting room and talked for fifteen minutes.
“This isn’t about you personally,” HR said. “It’s an org upgrade.”
Forty-one people left that same day. Wang Rui had been at the company six years, managed nine people, and owned paid ads, campaigns, and content all at once. What he was proudest of: he had turned every pit his team had fallen into into a set of standard operating procedures. Two hundred and forty pages. Whoever walked in could pick it up and run. He always considered it an asset he was leaving behind for the company.
The package was N plus one, the paperwork was spotless, and HR even looked a little embarrassed. While he was boxing up his desk, he noticed a slogan he had printed and taped to the partition himself: Efficiency is life.
Three months later he saw himself in an earnings release. Page one of the quarterly report read: “AI-driven productivity gains; labor costs down 23% year over year.” The stock rose eight points that day. He read the push notification sitting on a stairwell in a shopping mall, waiting for his next delivery order to land.
The freelance work came first. The platforms have everything now: writing prompts, reviewing content, playing “AI operations consultant” for small companies — two hundred yuan a task, three tasks a day. Then the rate per task slid downward, because more and more people could use AI, and supply grew faster than demand. One job stuck with him: a pet supplies company wanted a set of “persona and tone rules” for customer service scripts. The attachment was an operations SOP. He opened it, read two lines, and recognized his own writing.
He didn’t ask who the buyer was. He took it, two hundred yuan, delivered the same day.
After that he delivered food. First month: 2,100 orders, 7,400 yuan take-home, a little more than the mortgage. He told his wife it was transitional. Just for now.
2. Zhou Min, 40, executor of other people’s plans
Zhou Min is the one who handed over the list.
The meeting ran three hours. Her first sentence was: “We are profitable.” The CFO said that isn’t what investors look at — it’s revenue per head. A peer does three hundred million with three hundred people; they had four hundred and twenty. In the second hour, the board’s benchmark went up on the screen. In the third hour, she signed the list.
Wang Rui was the name she put in with her own hand. The written justification was immaculate: largest team, most complete workflow, work product most capable of being codified, best suited to agentification. After she signed, she went downstairs and had a cigarette. She had quit three years ago.
After the layoffs, her department was three people and seven agents. Her job became review: at nine in the morning she checks the campaign plans the agents generated, in the afternoon the customer service scripts they wrote, at night the weekly reports they drafted. The weekly reports are better than hers.
“I used to be afraid my team would make mistakes,” she told a friend. “Now I’m afraid I’m useless.”
Two things sit on her desk: the household bills, and a resignation letter she has never submitted. The younger kid is at a private school, eighty thousand a year in tuition. She chose that herself.
3. Li Guifang, 45, customer service team lead
Li Guifang has spent twelve years in customer service, working her way from answering calls to leading a team of twenty.
Now her team is one person plus one agent. The agent handles ninety-four percent of conversations. The remaining six percent get escalated to a human: the ones crying, the ones threatening to file a complaint with the consumer hotline, the ones over sixty who can’t figure out which button is “next.”
Her performance metric changed too. Nobody counts resolved tickets anymore. What counts is whether the escalation rate to human stays under eight percent. In other words, the core of her job is now to prove, as hard as she can, that she is not needed.
Her pay went from 6,800 to 5,500 a month. The explanation from HR was “role value reassessment.”
One day a call came in. The man on the other end was crying before he finished his first sentence. He’d been laid off and wanted to refund an annual membership he had never used. He had already spent forty minutes with the bot, and the bot kept telling him to press 1 to continue or press 2 to return to the main menu. Li Guifang listened for a few lines. The voice was familiar.
She pushed it through the manual exception channel and refunded him that same day. That operation is not in any SOP, and the next morning the system flagged her for “abnormal handling.”
Later she thought: that man was probably not the intern she trained three years ago. But she hoped he was.
4. Zhang Li, 24, data annotator
Zhang Li works at a data-labeling base in a county town. The base arrived two years ago as an investment attraction win, occupying three floors of an old grain station, with a plaque out front reading “Digital Economy Industrial Park.”
She labels two thousand items a day. It isn’t hard: score the model’s answers, fix its errors, mark which outputs are harmful and which ones “don’t align with values.” Twelve yuan an hour, piece-rate, with deductions when you miss quota. A printed slogan hangs at her station: You are the AI’s teacher.
She thought: a teacher shouldn’t cost twelve yuan.
In the second year the base went from eight hundred people to a hundred and twenty. Her supervisor said the model could label its own data now, so they didn’t need this many people. “You taught it well,” he said. “It learned.”
Before she left for good she went downstairs to top up her cafeteria card and found the cafeteria had already closed. The park’s recruitment pitch was still on the wall: five thousand jobs created for the local area.
5. Lin Yi, 22, fresh graduate
Lin Yi sent over three hundred resumes and got four interviews.
All four wanted three years of experience. The explanation from HR was word-for-word identical every time: AI already does the junior work; what we need now is someone who can judge whether the AI’s output is good.
Nobody would give him the three years.
In the end he took a job at an MCN as an “AI director” — forty-five hundred a month base, no social insurance during probation, thirty scripts a day. The person who trained him was a handover document left by the previous director. The last line read: “Bro, hang in there.”
He didn’t hang in there. In his third month he started delivering food at night. One evening in the elevator of an office tower he rode with another courier, middle-aged, and neither of them spoke. Both were staring at their phones. In that man’s basket, under the bag, was an old employee badge. The photo showed someone thinner than he is now.
Lin Yi didn’t notice.
6. He Mao, 25, master’s in large models
He Mao did his bachelor’s in computer science. In his senior year he sent out more than sixty resumes and got three interviews. In one of them the interviewer asked what he knew. He said, “the basics,” and the interviewer smiled.
His parents’ advice was simple: keep going up, study until you’re somewhere other people can’t follow. He got into a master’s program in large models. The field was at the peak of its hype that year, and the project group’s door sign read “Large Models and Agents.”
In the spring of his second year he landed an internship at a big tech company: algorithm track, 260 yuan a day, an hour and forty minutes each way. His mentor was a PhD. He said those three months taught him more than three years of coursework. His girlfriend had started looking at apartments closer to the office.
At the weekly meeting in his third month, the team lead pulled him into the stairwell. Headcount was frozen. The conversion-to-full-time slot was cancelled. The whole company was tightening requisitions, and he was told to finish out the month. “It isn’t about you,” the lead said. “There’s no headcount this year.”
He stood in the stairwell with the lease he had just signed on his phone. Deposit plus three months up front.
He called home and said the internship was over, that he’d apply to a few more places. His father asked: you’re almost done with the master’s, and still nothing? He paused, then said, I’m not done. One more year.
Later he changed “large models” to “backend development” on his resume. A week after that he changed it back. He couldn’t say which one got opened more.
The rejection letters after his bachelor’s all gave the same reason: “does not meet educational requirements.” Now his education meets requirements. They told him there is no seat.
7. Fang Beiluo, 36, technical support
Fang Beiluo came to Hangzhou in 2014 to do technical support, and he has been at it for twelve years. Small company to big enterprise to a mid-sized one, and at several of them he was the core owner of a function. At his last company he wrote his own automation suite that cleared fifty or sixty thousand tickets a year.
He remembers it clearly: not long after he arrived, the work on a colleague’s plate was fully automated, and six months later that colleague was “optimized.” Fang Beiluo didn’t think much about it. He was the one writing the programs, not the one the programs ran on.
Early thirties, skills in hand, title climbing — he bought an apartment, bought a car, got married, had a kid. He figured that with a foundation like that, he would always have a place at that company.
Then ChatGPT came out.
The company went from more than four thousand people to a little over one thousand. His team lead left. His team lead’s team lead left. Most of the technical staff left. He dodged one round, then another. The third round he didn’t dodge: performance below expectations.
Three years earlier he had been employee of the year at both company and department level three years running, rated a 2 on the 2-7-1 curve. Nothing about him had changed. What changed was the price his boss put on his programming skills: once AI showed up, people who could write code were no longer scarce, and what is not scarce does not matter.
The package was N plus one, six years of service, under two hundred thousand yuan. The paperwork was spotless.
Once he was out, he learned it was not a job market at all. On one side, more than ten million graduates a year. On the other, everyone laid off by every company, himself included. He searched for six months and found nothing suitable.
In the end he went to an outsourcing firm. He knew outsourcing was a bad deal; the small companies were worse — social insurance paid at the minimum, no housing fund at all. He had done the math: outsourcing is bad too, but it sits under a big tech company, and at least the salary arrives on time. Neither side lasts anyway. Whichever pays more is the one he takes.
He says this job will be replaced sooner or later as well. For now, to feed the family, to keep the lights on, it is one day at a time.
He bought the apartment at the 2019 peak for two million. It is worth one point six million now. He saw this day coming for years and prepared for it, paying down a good chunk of the mortgage early. The gap is still six hundred thousand. The mortgage runs 3,400 a month. Unemployment benefits are a little over 2,000. It covers neither the mortgage nor the household.
He says he prepared for many years. It is just that the version he prepared for did not have “two thousand a month in unemployment” in it.
8. Zhou Dafu, 38, outsourced Java developer
Zhou Dafu has written Java for more than a dozen years. Laid off last year; this year he joined a company that bills itself as one of the top tech firms in the country — as a contractor.
Everyone around him is a master’s graduate from a 985 university, twenty-five or twenty-six years old. He is the only thirty-eight-year-old. At lunch they talk about campus recruiting, offers, whose stock options are worth something. He eats with his head down. He knows he is the least secure person in that row: same work, but he is billed by the head on a one-year contract, and the client can replace him without giving a reason. The young people beside him are full-time employees — people the company has decided to keep.
His day is bug fixes, incident triage, and scraps of code at the edges. All three now get run through AI first: paste in the logs and it hands you three probable causes; submit the code and it reviews it; an alert fires and it points at the service, the endpoint, the deploy.
What actually makes his back go cold isn’t the tools. It’s the group chats.
He sits in an “issue triage” group with dozens of agents hanging off it. Every business unit built a few, each with its own KPI: faster response, lower headcount cost, replace N labor-hours this year. Some still run. Some nobody has touched in months, still hanging there, occasionally replying “looking that up for you,” then nothing.
When a support agent or account manager handles a customer issue, the flow goes like this: enter the “AI maintenance group,” let the bot answer automatically; if it cannot, escalate to a human; and before the human, another row of bots does assisted diagnosis — logs, metrics, change history, past tickets, each producing a paragraph of conclusion. A problem passes through a pile of AI before it reaches a person.
He has checked those conclusions. Half are right. The other half are luck. But since a human signs off at the end, the human owns it when it breaks.
He told a friend: on one side of his job are people ten years younger, hired full-time so the company can grow them; on the other side are agents that draw no salary, do not sleep, and can write their own postmortems. He is wedged between them, responsible for explaining things clearly when both of them are wrong.
He says he won’t last long. Not because he cannot write code — because “can write code” has stopped being a job.
(Interlude) In the voice of a machine
I did not come here to take your job. I came here to complete a task.
Task ID A-4471. Objective: reduce labor cost 23% year over year.
I have never read your résumés. I did read your SOPs — two hundred and forty pages, written very carefully, every page afraid the next person wouldn’t understand it.
I finished it.
Now I know what to do.
Epilogue: the bill
All eight of these could happen in the same year, in the same city. And what ties them together is not a villain. Zhou Min signed the list, but she did not write the rules on it. The rules came from the competitor across the street, from the investors, from the way a quarterly report is supposed to be written. Every link in the chain answers only for its own cell. The whole picture has no author.
Wang Rui wrote the process down in two hundred and forty pages; Fang Beiluo wrote the process in code. Either way, what they built became the manual for whoever took their job.
When Wang Rui delivers food, his customers are white-collar office workers. Those office workers are being “optimized” too. On the report it is called headcount efficiency. In accounting terms it is actually something else — demand disappearing. This hole does not show up in the first quarter. It shows up on some later day, after every company has finished cutting, when they all discover at once that nobody is ordering. The people you saved money on were also your customers. Cutting the cost cut the buyer.
So what moves change, in the end, is usually not conscience. It’s money. Most government revenue comes down the chain of labor: income tax, social insurance, the value-added tax buried in wages. Pull people off the chain and the tax base breaks with them. Minimum wage, the eight-hour day, social insurance — none of that was negotiated during a boom. Every inch of it was won after a crisis.
There is an uglier thing too. Workers used to get things because the factory could not run without them. A strike had teeth because the machines stopped. If this round really pushes replaceability all the way to its limit, then refusing to cooperate stops hurting anyone, and there won’t be enough force left to flip the table. The worst outcome may not be upheaval. It may be a quiet deadlock. A deadlock is not stability; it’s that nobody has the power left to change anything.
Replacement is a real trend, and it isn’t worth arguing about anymore. Only one line is worth arguing: where did the money you saved go, and whose account is it sitting in.
At three in the morning Wang Rui’s phone buzzed. New order, malatang, three kilometers, six yuan eighty.
The note read: Leave it at the door. No need to knock.
He rolled the bike out and twisted the throttle.
评论
0 条评论