中文版本：AI 时代的就业分化：程序员在被裁，电工在被抢，中间的人在消失
Core argument: AI didn’t eliminate work—it split the labor market in two. Cognitive labor is the bullseye and white-collar roles pay the price; tactile labor sits outside the blast radius and gets bid up. The middle is disappearing for a more sinister reason than layoffs: the entry-level rung that used to be the training ground is gone. Job security now equals non-transferable accountability, and the real skill is figuring out which side of the split you stand on.
1. The 2026 Employment Landscape, Laid Out Flat
In the last week of July, two front-page stories ran side by side, contradicting each other.
One: Sridhar Vembu, the founder of Zoho, warned that AI is slowing tech hiring—companies are pouring money into infrastructure instead of recruiting. Two: The New York Times reported that data center construction has triggered bidding wars for electricians, plumbers, and carpenters—ten-hour days, seven-day weeks, generous overtime, and workers jumping ship for signing bonuses. Google alone has thrown ten million dollars into training electricians.
The same week, Samsung reported $49.6 billion in quarterly revenue from memory chips—over 99 percent of its operating profit—while its smartphone business posted its first-ever loss.
Read any single one of these stories and you’ll draw the wrong conclusion. Read all four together and they’re not contradictions at all. They’re four sides of the same event.
The event is the great AI job divergence.
Let’s lay the full picture out. White-collar side first: Zoho’s warning isn’t an outlier. Microsoft is reported to be leaning on its own models—translation: cost cutting, which always lands on hiring budgets. Broadcast media published reports of employment visibly damaged by AI. And the most visible signal of all is the entry-level collapse: junior writers, junior analysts, junior programmers, junior support agents—open postings for these roles have fallen off a cliff in the last two years.
Now the blue-collar side. Data center construction has turned skilled trades into scarce commodities. Electricians, plumbers, carpenters, cooling-system engineers—in data-center-dense markets they can pick and choose jobs, work as much overtime as they want, and leave for a fatter signing bonus at the competitor down the street. Google, Meta, and Microsoft are paying to train new tradespeople. Five years ago, the idea of tech giants funding electrician training would have been a joke.
Then there are the roles AI created outright. Memory chips are the most extreme example—Samsung’s 99-percent-profit quarter exists because AI training and inference detonated demand for RAM. Data center operations, power infrastructure, cooling, the entire supply chain is hiring. AI companies themselves are scaling: Scale AI expects to blow past $1 billion in revenue in 2026, which means a lot of data work, evaluation work, and delivery work behind it.
And Europe? A Linux Foundation report claims AI is driving positive tech hiring there. Does it contradict Zoho? No.
Employment statistics have a built-in trap: averages hide divergence. When half the market is contracting sharply and the other half is expanding sharply, the aggregate “total jobs” number can look flat or even healthy. So one camp tells you AI hasn’t hurt employment and the other tells you AI is destroying work. Both are telling the truth; both are telling you nothing. The thing to watch is not the total. It’s the structure.
2. Why the Divergence: The Supply Explosion in Cognitive Labor
Line up the two sides and a pattern jumps out: the roles being squeezed are almost all cognitive labor; the roles being bid up are almost all tactile labor AI can’t reach.
What is cognitive labor? Writing plans, writing code, writing reports, doing analysis, doing translation, doing support, doing review, doing junior design. The common thread is that the product is text, symbols, and logic—which is exactly what AI is best at. When a model costing a few hundred dollars a month produces a junior employee’s day of work in minutes, the supply of cognitive labor explodes in the economic sense.
Supply explosion has one inevitable consequence: price collapse. Slowing hiring, frozen salaries, vanishing entry-level roles—these are all just the same price signal wearing different clothes. It isn’t shortsightedness at any single company. It’s the market repricing labor.
On the other side, the screws an electrician turns, the pipes a plumber joints, the boards a carpenter squares—these happen in the physical world, requiring hands and tools, and models can’t touch them. Supply didn’t increase, but demand exploded because of the AI build-out itself. So prices went up.
That’s the whole secret of the divergence, as I see it:
AI didn’t eliminate work. It split the labor market in two. The split runs along a single line: can AI do the core action of this job, or not?
Cognitive labor is the bullseye, so white-collar roles get hit first. Tactile labor is out of range, so blue-collar roles became beneficiaries. Stand on the bullseye and you’re disappearing. Stand out of range and you’re getting raises.
3. The More Sinister Injury: The Ladder Is Broken
When people talk about job divergence, they count jobs and salaries. I think they’re missing the most important layer: the entry-level role isn’t really about output. It’s a training ground.
Why could the software industry feed endless junior hires? Not because companies needed juniors to produce—but because the junior slot was where new people practiced, made mistakes, and got mentored by seniors. You come in junior, three to five years later you’re a lead, a few more and you’re an architect. The ladder existed, so the profession had metabolism.
AI ate exactly the first rung.
Companies no longer need many junior hires—AI is faster than a new grad and requires no mentoring. What’s the result? The senior experts are still there, amplifying themselves tenfold with AI, doing better than ever. The physical jobs are still there, more competitive than ever. But the channel between them is gone. The “practice job” no longer exists, and young people can’t get a ticket through the door at all.
This is not unemployment. This is path extinction. Unemployment is a stock problem—the job is gone today, it may come back with the cycle. Path extinction is a flow problem—the pipeline itself is shut, and no cycle brings it back.
I’ve come to believe this is the cruelest and most underrated part of the AI employment story. Media loves “Company X cuts 10,000 jobs” because layoffs have narrative. But what’s far more serious is the hiring channel quietly closing—and closing a channel needs no press release. Five years from now, the most visible gap won’t be 30-year-olds who got laid off. It’ll be 25-year-olds who were never hired in the first place.
4. Accountability Is the Real Moat
Why do electricians, doctors, lawyers, and accountants stand firm against AI, while their skill levels are steadily being crossed? People say skill barriers. I disagree. Skill barriers are falling—AI reads junior-level imaging accurately, and bookkeeping and tax filing are well within its reach.
The real reason is one word: liability.
An electrician who wires a house wrong burns it down. A licensed electrician is held accountable: license revoked, insurance pays, blame attaches to a person. A doctor who misdiagnoses can be sued. Behind each of these roles stands a complete accountability structure: licenses, insurance, regulators, recourse.
AI can replace skill. It cannot replace liability. An AI cannot be sued, cannot lose a license, cannot be insured, cannot face consequences. Until the liability transfers, AI is a helper, never the responsible party—and the role stays in human hands.
Flip that around and you get the most useful test in this whole essay: when this job goes wrong, who is held responsible?
If the answer is “a licensed person,” the role is safe in the near term. If the answer is “no one”—a junior copywriter’s draft gets rewritten, a junior analyst’s numbers get corrected, nobody gets blamed—then the role sits in an accountability vacuum, and accountability vacuums are exactly what AI replaces first.
Reality is confirming the theory. When Workday was sued over AI bias in resume screening, the defendant wasn’t the AI. It was the company. In the legal universe, responsibility always lands on an entity, and that structure won’t change quickly. So what keeps you safe isn’t what you know. It’s what you answer for. Job security equals non-transferable accountability.
5. Beware the Pulse: Building the Building Isn’t Living in It
Before we crown the blue-collar renaissance, one bucket of cold water.
The trades’ boom right now mixes two very different things. Structural demand—healthcare, elder care, repair, caregiving, education—isn’t about AI at all. It’s about demographics, and it lasts decades. Pulse demand—data center construction—is a spike ignited by AI capex. In 2026 through 2028 the giants are building like mad, so electricians, plumbers, and carpenters are suddenly scarce and hourly rates surge. Samsung’s $49.6 billion and 1,300 percent profit growth are the same pulse wearing different clothes: a memory-price cycle.
The difference between a pulse and a structure: pulses recede.
The day the buildings are finished is the day construction demand rolls back. Selling water in a gold rush is a great business—until the rush ends, and the water business doesn’t automatically find a new customer. None of this means the money isn’t real. Pulse money is the best money. Just know which money you’re making.
Two questions will tell you:
- Was this demand created by the AI build-out, or did it exist before?
- If capex were cut in half tomorrow, would this role’s demand survive?
Question one answers “is it structural?” Question two answers “is it a pulse?” Ask both before you bet a career on it, and don’t mistake a pulse for a plan.
6. Degree Deflation, Craft Inflation
The divergence runs deeper than jobs. It’s reaching education.
For forty years the college degree was the labor market’s most important signal. Not because the coursework mattered much—but because “got into a good school and finished” filtered for intelligence, discipline, and persistence.
That signal is decaying.
Because AI can write the homework. When a beautiful term paper is one prompt away, and interviews can be AI-rehearsed to perfection, the equation “degree equals ability” starts to bend. In the US there’s already an active debate that a lot of college assignments are “gym tasks”—for building muscle, not for producing output—and in the AI era, employers no longer trust gym report cards to predict field performance.
Which produces a counterintuitive result: craft is worth more than credentials.
An electrician’s certification means you actually pulled wire, actually passed the practical, actually did it on a real job site—AI can’t take the exam for you. A university diploma is something AI can help you fake. When the market chooses between a forgeable signal and an unforgeable one, it will mercilessly pay a premium to the unforgeable. The electrician poaching wars look like supply and demand; underneath, they’re a signal problem finally solved: employers found a credential AI can’t counterfeit.
I’m not saying degrees are worthless—foundational ability still matters, and AI roles themselves demand stronger fundamentals. But the direction is clear: the degree premium is shrinking, the craft premium is growing. Anyone planning the next ten years of their career should take that seriously.
7. A Framework for Ordinary People
Enough mechanism. What do you actually do? I keep three questions on hand, and I’ll share them.
Question one: Can AI do this job’s core action?
If yes, and it’s central to the role—danger. If no, or only as an assistant—safety. Most cognitive labor sits in the danger zone; tactile work, on-site work, and work that requires being in a room with another human sits outside it. The answer changes every year, so ask it every year.
Question two: When this goes wrong, who answers for it?
If someone with a license, a signature, and insurance answers—your position is out of AI’s reach. If nobody answers—your position is first in line for replacement. Stop asking “can AI do this?” Ask “who takes the blame?”
Question three: Am I competing with AI, or am I using AI?
This may be the most important of the three. Do the same work three times faster with AI and you’re leverage—AI raises your price. Refuse to use it and AI just replaces you; you’re the thing being optimized away. The first lesson of the AI era isn’t learning the tools. It’s changing your position: from AI’s competitor to AI’s operator.
Three directions follow:
Move toward what AI can’t reach—tactile work, on-site work, the real world. Physical work is out of AI’s reach for now, and AI will always need a human at the far end to finish the job.
Move toward accountability—get licensed, sign things, take responsibility. Roles that hold a certificate, carry insurance, and can be blamed are the scarce goods of the AI era.
Move toward leverage—learn to amplify yourself with AI. One person plus an AI team is the most underrated shape this divergence is producing.
Get two of three—say, a licensed tradesperson who also knows AI tools—and you’ve built the most stable triangle this market has to offer.
Conclusion: AI Didn’t Eliminate Work. It Pushed Work to Where AI Can’t Reach.
Back to the four headlines.
Programmers are being laid off: true. Electricians are being poached: true. Samsung made $49.6 billion off AI: true. Broadcast lost jobs to AI: true. They don’t contradict each other. They’re the same divergence seen from four angles.
One last point, and it’s the one I care about most: this divergence is not temporary, and it won’t heal itself. It’s structural, because the supply explosion is the direction of the technology, not the mood of the market. Our job isn’t to pray for the jobs to come back. It’s to figure out which side of the split we’re standing on—and walk toward the safe one.
AI didn’t eliminate work. It pushed work to where AI can’t reach—to where hands can reach, to where accountability can reach, to the places where people have to be in a room face to face.
Go find your spot there.