Three Questions After the AI Job Wave

The drive to hire is weak right now.

Before, when you wanted to build something bigger, your gut reaction was "we need more people." Now it's the opposite: "how do we squeeze more out of who we already have?" Even bringing on interns feels less appealing.

It's a small shift in sentiment, but it points to something that isn't reversing.

Two days ago, Kim Yong-beom, policy chief at the South Korean presidential office, posted on Facebook: the "excess returns" of the AI era shouldn't belong only to individual companies; part should flow back to the public as a "citizen dividend." The next day, the KOSPI plunged 5.1%. He later clarified he wasn't suggesting confiscating profits, only discussing how to spend the "excess tax revenue" created by AI dividends. The market settled.

A single Facebook post moving the market 5% tells you the issue is already on the table.

The context is plain enough. SK hynix posted a 72% operating profit margin in the first quarter; spread across employees, bonuses averaged nearly $500,000 per person. Samsung Electronics' semiconductor division logged 53.7 trillion won in operating profit for the same period, but the 74,000 workers represented by the union received a far smaller slice than their SK counterparts. The Samsung union has threatened an 18-day strike starting May 21. The workers aren't after a slightly larger bonus. They want a respectable cut of the AI dividend chain.

I see this as the middle of three questions AI is really throwing at society. Ahead of it is the unemployment question. Behind it is a deeper one about identity. The three are linked.

Question One: Net Job Loss Is the Pattern

First, let's get the definitions straight.

Mainstream reports don't actually agree on "global net job losses." The World Economic Forum's Future of Jobs Report 2025 still predicts a net increase of 78 million jobs by 2030. The ILO's GenAI exposure index uses "transformation" rather than "replacement": one-quarter of jobs globally are touched by GenAI, one-third in high-income countries. The IMF uses the broadest brush: 40% globally, 60% in advanced economies.

But macro models are neat; actual labor pools are not. When a clerk gets "transformed" into an "AI-collaborating high-efficiency role," the macro report counts that as transformation. For the clerk, it's unemployment. New jobs like AI product managers, data governance specialists, model evaluators, and robot maintenance technicians may not be in the same city, may not suit the same age group, and may not absorb the former customer service reps, copywriters, junior programmers, and admin staff. Someone always gets pushed off along the way.

The current data already shows how painful this transition is. In Challenger's March 2026 U.S. layoff report, AI was the top reason given for cuts: 15,341 people, 25% of all monthly layoffs. Tech layoffs in the U.S. reached 52,050 in Q1, up 40% year over year. Goldman Sachs estimates that over the past year AI has eliminated roughly 25,000 jobs a month while creating fewer than 9,000. Net loss: 16,000. Gen Z and entry-level white-collar workers are hit hardest. Research from Stanford Digital Economy Lab also shows that in the jobs most exposed to AI, employment among 22- to 25-year-olds has dropped markedly.

This is what I mean by "net loss." I'm not forecasting global employment in 2030; I'm looking at the real demand for specific occupations, specific age groups, and specific companies right now. When a company realizes that ten people plus AI can do what used to take fifteen, it doesn't first ask a macro model whether new jobs will appear in five years. It freezes hiring, trims headcount, and cuts peripheral roles.

The sectors that traditionally soaked up labor are running in reverse, too. Autonomous vehicles are replacing drivers; drones are replacing delivery riders; industrial robots are replacing line workers. IFR World Robotics data shows global manufacturing robot density doubled in seven years. In China in 2023 there were 470 robots per 10,000 manufacturing employees. New infrastructure no longer naturally brings large numbers of low-barrier jobs the way building bridges and roads once did. Computing centers, ultra-high-voltage grids, battery plants, and dark factories are all capital-intensive and light on labor.

Some pin their hopes on "one-person companies." OPCs have been hyped plenty over the last couple of years, and I do think they'll become a real new organizational form. By June 2025, China had over 16 million one-person limited liability companies; 2.86 million were newly registered in the first half of 2025 alone, up 47% year over year. Shangcheng District in Hangzhou is already piloting OPC community policies.

But judging whether OPCs can carry employment means looking past the headline numbers and anecdotes. Most of those 16 million are traditional self-employed operations and micro-entities that existed long before AI. Two things matter: the growth rate, and what share of that growth can actually sustain middle-class incomes.

The growth rate is eye-catching: 47% year over year. But the distribution is ugly. Industry reports show OPC revenue is extremely long-tail: more than half are still stuck in a product-validation phase earning a few thousand yuan a month; fewer than one in ten steadily clear a million yuan a year. Even with an optimistic 10%, only 600,000 of the 6 million new OPCs each year would reach middle-class levels.

China's 2025 statistical bulletin puts year-end employment at 725 million. Apply the IMF's exposure metric: 60% in advanced economies. Use a more conservative 30% for China, and that's still over 200 million people. 600,000 versus 200 million: two orders of magnitude apart. OPCs will buoy some super-individuals, but they can't hold up the labor market.

Why is this shock so sharp? I boil it down to one word: concentration.

SK hynix posted 37.6 trillion won in Q1 operating profit with roughly 35,000 employees company-wide. That's roughly 1 billion won in operating profit per employee for the quarter, over 4 billion won annualized, or more than 20 million RMB per person. Not every employee actually creates that much, but the number makes it viscerally clear how few hands the AI dividend is squeezed into.

Xiaomi isn't as extreme, but the direction is the same. In 2025 the group recorded 457.3 billion yuan in revenue; its smart EV and AI innovation business contributed 106.1 billion yuan at a 24.3% gross margin, delivering 410,000 vehicles for the year. Automakers used to compete on production capacity; now they also compete on algorithms, supply-chain software, automated production lines, and data loops. Capacity is scaling up; headcount isn't scaling with it.

Since the Industrial Revolution, every major industrial wave has pulled new job chains along with it. Labor scale and industrial scale mostly moved together. This AI wave runs the other way: the greater the output, the fewer people it needs. Chips, cloud, models, data centers, plus the handful of teams that can push AI to its limits. These swallow most of the dividends.

You can think of it this way: one person out of a thousand, armed with super-productivity, flattens part of the work that the other 999 used to do. Not all jobs are erased, but the demand curve flattens. I don't see any new direction that could regenerate labor demand on that scale in the same window.

Accepting this is prerequisite to discussing the next two questions. Net job loss isn't a panic slogan; it's a magnitude mismatch already happening in local labor markets.

Question Two: Distribution Needs a Reset

Which brings us to the second question: distribution.

The Korean incident is a template for the distribution question. Once AI dividends concentrate in a handful of companies, who gets them? Shareholders, executives, core engineers, all employees, or society at large through taxes and public spending? This question will inevitably spread from Korea to Japan, Europe, and China.

If concentrated dividends are still distributed under old rules, the outcome is almost predetermined. The bulk goes to shareholders and top management; core employees get hefty bonuses; ordinary workers get fixed salaries; and most people outside the supply chain only feel rising prices, rents, fewer openings, and tougher competition. This structure was already widening inequality; AI only steepens the curve.

The group squeezed hardest isn't the lowest earners. It's the middle class, especially those living on salaries, professional skills, and stable jobs.

The reason is simple. Wages are the most perfectly taxed form of income; there's nowhere to hide. In China's individual income tax system, comprehensive income faces progressive rates from 3% to 45%. Salaries are withheld at source, and social insurance plus tax are deducted before the money ever hits your hands. High-salary workers should certainly pay tax. The problem is that wealthier people have far more types of income: capital gains, equity incentives, dividends, corporate structures, trusts, family offices, cross-border arrangements. I'm not saying these are illegal. I'm saying they have more choices of tax base and more room to defer.

This structure was already obvious in the internet era; in the AI era it will only get worse, because the core gains of the industry concentrate in fewer hands. The EU Tax Observatory's global tax evasion report also notes that billionaires worldwide face extremely low effective tax rates relative to their wealth. The more mobile wealth becomes across borders, the harder it is for any single country to carry out redistribution alone.

The other side of the middle-class squeeze is how fast they are being replaced. The jobs AI currently hits easiest are middle-class occupations: programmers, designers, customer service reps, junior legal staff, junior analysts, copywriters, translators, operations specialists, admin staff, finance assistants. They shoulder the heaviest taxes and face the fastest replacement. They are the ones hurting most in the current structure.

Back to Korea. The Samsung and SK unions aren't fighting over a one-time bonus; they're fighting over a long-term rule. The companies will only offer a "special bonus." The unions want the profit-sharing ratio locked into a formal agreement that takes effect every year. On the surface it's about the bonus amount. In reality it's about whether this distribution rule will still hold next year.

Using "excess profits" or "excess tax revenue" for redistribution isn't a new framework in itself. Nordic countries have been running this for decades. Denmark's top marginal income tax rate is pushed to 60.5% in 2026. Sweden, Finland, and Norway have long maintained high labor-tax burdens and public services. The OECD's Taxing Wages also shows that the average tax wedge on labor in European countries is markedly higher than in the U.S. or Korea.

But the AI era introduces a new problem: productivity itself can move.

Heavy-asset, fab-heavy players like Samsung and SK hynix can't move; the Korean government can at least capture some corporate income tax and supply-chain revenue. But the more typical AI business doesn't look like that. Compute is rented in Singapore; the company is registered in Ireland; the team is spread across five time zones; settlements run through global payment networks. Teams of three to five people generating hundreds of millions in revenue will become more common, and nations have far fewer levers to tax them than they had with traditional manufacturing.

So an "AI tax" can't be read as simply slapping higher taxes on a few companies. It's more like a bundle of questions. What is the tax base? Compute, profits, capital gains, data, or the labor costs displaced by robots? And who receives the revenue? Is it poured into new infrastructure, or used to shore up social security, education, health care, pensions, unemployment insurance, even direct cash transfers to residents?

What needs guarding against here is path dependence. Many countries have grown used to propping up the economy with investment and infrastructure, but AI-era infrastructure may not prop up employment. Building more computing centers, data centers, ultra-high-voltage grids, and battery plants will likely continue to raise the productivity of leading firms, benefiting capital and a narrow slice of high-skill jobs, while doing little directly for displaced middle-class and low-income workers.

This is why people at OpenAI have been talking publicly about UBI for years. OpenResearch, funded by Sam Altman, ran a three-year experiment in Texas and Illinois: 1,000 low-income participants received $1,000 per month, alongside a control group of 2,000. The results, published in 2024, weren't miraculous. Recipients worked an average of 1.3 fewer hours per week, had a 2 percentage-point lower employment probability, and saw household income excluding subsidies decline. But they were more proactive in looking for work, valued meaningful work more, had more room to relocate, see doctors, and plan long term, and were more likely to have entrepreneurial ideas.

This experiment matters, not because it proves UBI is right, but because it drags the debate from slogans back to data. Cash doesn't automatically make people stop working, nor does it automatically give them dignity. What it provides is a buffer and choice. For a society with excess productivity and rapid job restructuring, choice itself may be infrastructure.

I don't think UBI is the standard answer for the AI era. But it's one of the few options that has been seriously tested and has data behind it. Compared with patching old rules, it at least offers a different starting point.

Question Three: Where Does Value Come From for People Who Don't Work?

This is the hardest of the three. The first two can still be moved forward with policy, tax systems, and redistribution. This one cannot.

In the Chinese context, "not working" is a very heavy verdict. At family gatherings, when someone asks "What do you do?" the expected answer is an occupation. If you reply, "I don't currently have a job," the atmosphere changes instantly. This isn't just about face.

The National Bureau of Statistics' 2025 bulletin lists 5.95 million urban residents and 33.4 million rural residents on subsistence allowances at year-end. China does have a welfare system. But subsistence allowances and relief still carry stigma in many places. Families who qualify but don't apply have always existed. The reason isn't insufficient money; it's the fear of being whispered about for "living off handouts."

This sense of shame runs deep. Our generation grew up on a narrative that said "Work hard and you'll be rewarded; effort deserves respect." Education, media, and the people around you all tell you the same thing: your value equals your output. Labor is the anchor of identity; salary is the measure of it. I wrote a piece on AI anxiety before, touching on the other side of this. When AI fortune-telling trends and young people flock to mysticism for certainty, what's really happening is that this anchor is loosening.

AI has simply used up the expiration date of this narrative ahead of schedule. What it truly shakes isn't just income. It's the sense of identity. You receive a basic income, food and housing are covered, friends respect you, but you wake up with nothing to look forward to. That hollowness is something policy cannot answer.

A society whose time has been freed by AI doesn't lack welfare distribution points. It lacks a narrative that can give people a new identity. This isn't something engineers can code or models can compute. It requires telling a new story about what kind of person is worthy of respect and what kind of life is decent.

A thousand years ago the story was "study to become an official"; a hundred years ago it was "industry saves the nation"; thirty years ago it was "go into business." Over the last decade or so it was "get into a big tech firm," "buy an apartment," "start a company," "financial freedom." What it should be in the AI era, no one can give a clean answer.

Closing

The three questions are linked. The first makes the second urgent. No matter how well the second is handled, the third cannot be avoided.

The market tremor triggered by that May 11 proposal in Korea is only the opening act. I expect these discussions to spread to Japan, Europe, and China in the second half of the year. Every country will craft different answers based on its own politics and culture. AI taxes, UBI, tax-base reforms, new infrastructure, promoting one-person companies—each will have its trial runs. Trial and error itself is part of the answer.

What individuals can actually do isn't complicated: build more skills, keep more capital on hand, and don't let any single narrative sweep away your emotions. What society must do is harder: stop brushing things off with "new jobs will always appear," and stop pushing the unemployed back into shame. No one can answer all three questions at once. We can only walk through them one by one.

Three Questions After the AI Job Wave

Question One: Net Job Loss Is the Pattern

Question Two: Distribution Needs a Reset

Question Three: Where Does Value Come From for People Who Don't Work?

Closing

References

Recommended Reading

Subscribe to Updates

Comments