1. The Man from Asan
Friday, March 15, 2024. Baebang-eup, Asan, South Chungcheong Province.
At the main gate of Hyundai Motor's Asan plant, Lee Jung-hoon — 53 years old — handed back the employee badge he had carried for 28 years. He had joined the company in 1996, the year he graduated in mechanical engineering from Inha University. On the first day of his career, he was the youngest worker on the production line. On the last day, he was the senior department head of production technology.
Lee Jung-hoon's expertise lived in his hands. He could assess the alignment of a press mold from the sound it made on the line. He detected defective paint from the faint difference in sheen across a part's surface. He rose to department head because he led a process-improvement project that brought the defect rate down from 0.3 percent to 0.09 percent. His 20-odd team members called him not "section chief" but Hyung — older brother. Lunch was always in the company cafeteria. Some evenings he ate samgyeopsal with his team in Asan's new-town district. On weekends he played badminton with his elementary-school daughter.
That ordinary life began to fracture in January 2023. A new plant director arrived and announced the introduction of an AI-based quality-prediction system. Machine-vision cameras were mounted above the production line. IoT sensors collected equipment operating data in real time. Lee thought at first that these were supplementary tools. He believed that 28 years of accumulated sound, smell, and touch could not be replaced by AI.
Nine months later, the defect rate had fallen further. AI had done it. Lee's team became redundant. The company did not call this displacement. It called it "role realignment." His new title was "AI System Monitoring Supervisor" — a position that meant watching all day as monitors displayed results AI had already processed. The substantive work amounted to 5 percent of what he had done for 28 years.
A voluntary-retirement package was offered. Severance of ₩240 million — roughly $174,000. Lee deliberated for three days and accepted. He used half that sum plus a ₩50 million loan to open a chicken franchise. As of 2025, monthly net profit stands at ₩1.5 million. Less than what a part-time worker earns working six hours a day at minimum wage.
Lee Jung-hoon is not angry. That is the saddest part. He believes he was "left behind by the era." To his daughter he explains: "It wasn't that Dad didn't work hard — the world changed." Yet even as he says the words, he cannot quite convince himself. The hardest thing for him is not the feeling of incapability. It is the feeling of uselessness. That when AI makes an error, he cannot catch it. That his 28 years left the factory without being translated into data.
In Book 1, The Displaced and the Discerning, we saw the pattern by which the Hanseatic League's monopoly was bypassed through technological innovation. When Dutch merchants broke the Hansa's herring monopoly with the technique of salting at sea, the Hansa's skilled workers lost their purpose. Lee Jung-hoon did not know he was one of those skilled workers. No one had taught him why the Gracchi brothers failed.
2. What the Machine Learned
This is not the story of Lee Jung-hoon alone. Nor are the four individuals in this chapter a statistically representative sample. But the pattern repeated across their stories — technological transition, capital displacement, individual drift — aligns precisely with what the macro data shows. The whole of Korean manufacturing stands on the same formula.
Manufacturing employment in Korea peaked at 4.67 million in 2016 and has been in long-term decline since. As of February 2025: 4.386 million. The sector's share of total employment has fallen from 20.4 percent in 2000 to 15 percent. Workers enrolled in manufacturing employment insurance have posted six consecutive months of decline. But these numbers are only the tip of the iceberg.
A quieter collapse is underway. Process managers, quality team leaders, production-planning deputies — the intermediate management layer that connected the shop floor to senior leadership is disappearing. The domains AI replaces first, and most efficiently, overlap exactly with what these people do.
Consider process management. Real-time IoT sensor data and machine-learning anomaly-detection systems now perform the integrated judgments — on equipment utilization rates, defect patterns, worker skill levels — that experienced process team leaders once made by hand. Predictive maintenance — the technology that forecasts machine failure before it occurs — has replaced the process team leader's sense with data. Quality inspection has followed the same path. Machine-vision cameras have taken the seats of staff who once interpreted statistical process-control data and coached supplier quality. In Samsung SDS smart-factory cases, AI manages everything from lot tracking to supplier shipment quality to inspection records at a single auto-parts manufacturer. Production planning changed even faster. Cloud-based APS — advanced planning systems — now process in minutes and push automatically to shop-floor terminals the work that deputies once coordinated by hand: matching incoming sales orders against inventory and equipment capacity.
Samsung Electronics has declared that it will convert 100 percent of semiconductor production to AI factories by 2030. The government plans to expand AI-integrated smart manufacturing sites from 26 in 2024 to more than 100 by 2030. The OECD has assessed that 38.8 percent of current Korean jobs face more than 70 percent automation risk. Low-skilled manufacturing employment is projected to fall approximately 10.2 percent over the next decade.
Yet the label "low-skilled" creates a false impression. The tacit knowledge Lee Jung-hoon accumulated over 28 years — reading mold condition from sound, detecting lubricant failure from smell, sensing surface texture by touch — is classified as "low-skilled" by educational criteria, but the complexity of those judgments is anything but low. AI absorbs precisely that tacit knowledge by converting it into data. In Chapter 3 we saw Lee Kuan Yew design the fourth condition of indispensability — the reproduction of human capital. That Lee Jung-hoon's 28 years of experience left the factory without being translated into data means that human capital was not reproduced. It was discarded.
In Book 1 we followed the Lancashire handloom weavers as machines took their hand skills. The weaver knew the machine was replicating his hands. Today's plant manager watches AI replicate his judgment. The difference is scale and speed. The displacement of the handloom weavers took 30 years. Lee Jung-hoon's displacement took fewer than three. AI is faster than the steam engine.
3. A Stamp That Approves
Kim Su-jin is 44. Deputy director of corporate banking at KB Kookmin Bank's Gangnam branch. She joined the bank in 2004 — 20 years as a loan examiner. She graduated in business administration from Yonsei University and passed the CFA Level I. The ability she takes the most pride in is reading the person behind the documents. If the numbers in a financial statement were strong but the owner's eyes looked uncertain, she put the application on hold. If the numbers were weak but the business history was solid, she approved it. In 2018, a small food-ingredient supplier she approved for a loan was listed on KOSDAQ three years later. That moment gave her the greatest sense of meaning she has ever found in her work.
Bill-of-exchange brokers gave way to bank loan officers. Loan officers are giving way to credit-scoring algorithms. The gatekeeper of capital allocation has always been replaced by technological innovation. Kim Su-jin is the last generation of those gatekeepers.
In 2022, the bank introduced AI Credit Scoring System 2.0. At first, it was a reference tool. When the system issued an "approval recommendation," Kim made the final call. But overriding the system's recommendation grew progressively harder. If she approved a loan the AI had flagged for rejection and the borrower subsequently defaulted, the outcome was recorded as a "departure from examination standards."
In 2023, the performance evaluation criteria changed. The match rate between AI recommendations and the examiner's decisions was added as an evaluation item. A match rate above 85 percent earned an "excellent" rating; below 80 percent triggered a "needs improvement" designation. Kim began measuring, by her own hand, how differently she judged things from the AI. And that "difference" began counting against her score.
By 2025, her team's loan examiner headcount had fallen from four to two. Vacated positions simply were not refilled. The bank called it "natural attrition." Kim Su-jin still holds her seat. But what she does has become unclear. Lee Jung-hoon was displaced. Kim Su-jin was not displaced — instead, she was rendered useless. Not replacement but obsolescence. Still present, role dissolved. She became a stamp that ratifies the AI's decisions.
What she fears most is not being wrong. It is being right without being able to prove it. She once approved a loan for a company the AI had rejected, and that company grew. But that outcome was recorded only as "a lucky success following departure from examination standards" — not as evidence of her judgment.
The numbers confirm the structural shift. Over five years, 1,189 bank branches closed. The combined branch count of the five major commercial banks stood at 3,748 as of end-2025 — a sixth consecutive year of decline. In 2024 alone, 110 branches disappeared. The headcount of the four major commercial banks fell by 1,478 — 2.7 percent — in three years. The absolute figures appear modest, but accounting for voluntary-retirement departures and hiring freezes, the real contraction is far larger.
Internet-only banks are accelerating the pressure. K bank's cost-to-income ratio stands at 29 percent — the lowest in the industry and the only figure in the 20s. KakaoBank serves 26.7 million customers with a workforce of a few thousand. A traditional bank branch averages 15 to 20 staff, and there are thousands of such branches. The efficiency gap is structural. In 2024, all three internet-only banks posted record profits simultaneously. The traditional bank's only way to close that gap is to shutter more branches and replace more staff with AI.
Shinhan Bank has placed approximately 150 digital-desk terminals equipped with AI bankers in its branches and opened unmanned "AI Branch" locations offering 64 teller functions with AI alone. Financial regulators announced a 2024 plan to support generative AI adoption in financial services — permitting commercial AI services on internet networks and building finance-specific AI training data. Further decline in call-center and teller roles is already announced.
The gatekeeper of 40 years ago is disappearing now.
4. The Chicken Shop as Final Destination
Return to Lee Jung-hoon's chicken shop.
Korea's self-employment rate is 23.5 percent. The OECD average is approximately 15 percent — more than twice the rate of the United States and Japan. As of April 2025, 5.615 million self-employed workers. That number itself represents structural excess. And the way that excess resolves is not through new businesses opening — it is through businesses closing.
In 2024, self-employment closures in Korea surpassed one million for the first time on annual record: 1,008,282 closures — the highest figure since statistics began in 1995. A closure rate of 9.04 percent, the highest since 2020. Forty-four percent of self-employed workers said they were "considering closure within three years." Retail and food service account for 45 percent of all closures.
Survival-rate statistics are colder still. The three-year business survival rate is 53.8 percent; the five-year rate is 39.6 percent. The average lifespan of a restaurant is five years and one month — more than three years shorter than the all-industry average of eight years and four months. Chicken shops are worse. Three-year survival rate: 45.4 percent. Average startup cost: ₩93.94 million — roughly $68,000. Invest close to ₩100 million, and nearly half close within three years.
The pattern runs like this. A middle-class man who takes early retirement from manufacturing or finance opens a franchise with his severance package and a loan. A chicken shop, a café, a convenience store. Average investment: roughly ₩100 million. Closed within two to three years.
The severance package is consumed. The time it takes for 30 years of accumulated workplace savings to evaporate is fewer than three years. Financial assets excluding real estate converge toward zero. In August 2025, the Korea Economic Daily reported: "Thinking you can just open a chicken shop when you retire — that no longer works." The last safety net no longer holds.
Lee Jung-hoon stands squarely on that pattern.
Platforms press the final air out of self-employed workers. Delivery app revenue accounts for 75.7 percent of total chicken-shop sales — three-quarters of revenue passes through a delivery platform. Baemin's intermediary fee rose to a maximum of 9.8 percent in July 2024, and a 6.8 percent fee on packaging orders that had previously been free was introduced. Combined intermediary, delivery, and advertising fees consume 24 percent of total revenue. If Lee's chicken shop generates ₩6 million in monthly revenue, ₩1.44 million of it goes to the platform. After subtracting ingredient and rental costs, ₩1.5 million remains. That is why the number is ₩1.5 million.
Baemin holds 42.6 percent of the market; Coupang Eats holds 42.1 percent — effectively a duopoly controls distribution. In Chapter 13 we saw the chaebol's concentration structure simultaneously producing Korea's economic efficiency and its vulnerability. Platforms carry the same structure. The algorithm sells exposure, collects fees, and monopolizes data. The self-employed worker handles production; the platform controls distribution. Owning the means of production is not the same as controlling the distribution network.
Just as Rome's smallholders owned their land yet lost to the competition of large-scale slave agriculture, Lee Jung-hoon owns his shop yet cannot escape the control of platform algorithms. Just as the Gracchi brothers attempted land reform in Rome and failed, the Korean government discusses a fee ceiling for delivery platforms — but legislation remains incomplete. Regardless of whether one owns the means of production, whoever controls the distribution network holds economic sovereignty.
Was there another path for Lee Jung-hoon? If a role had existed to translate his 28 years of press-process experience — reading mold condition from sound, detecting lubricant failure from smell — into training data for AI systems. If a route had existed to join a manufacturing AI startup in Pangyo as a "domain expert." But that path did not exist in March 2024 when Lee retired. The only door standing open was a "Python and Big Data" course on the third floor of the local Employment Welfare Plus Center. No one had seriously designed where that door could take Lee Jung-hoon.
5. The Credential Paradox
Lee Jung-hoon's daughter is in the second year of middle school. Monthly private tutoring costs: ₩800,000. More than half of the ₩1.5 million net profit from Lee's chicken shop. He says he cannot cut it. "At minimum, she has to get into university."
The numbers reveal the other side of that belief. Korea's university enrollment rate exceeds 76 percent — one of the highest in the OECD. The country has more than 400 four-year universities and junior colleges. Total private tutoring expenditure in 2024: ₩29.2 trillion — approximately $21.2 billion. A fourth consecutive all-time high. Student numbers are falling while tutoring costs are rising. Monthly average per student: ₩474,000. Households earning above ₩8 million per month spend ₩676,000; those earning below ₩3 million spend ₩205,000. Among all OECD countries, Korea ranks first in the private share of total education funding at 37.2 percent — more than twice the average of 16.1 percent that households elsewhere bear.
And yet: the unemployment rate for high-school graduates is 5.5 percent. For university graduates, it is 6.7 percent. Higher education correlates with higher unemployment. The share of university graduates among economically inactive young people rose from 19.4 percent in 2015 to 23.7 percent in 2024. More than 400 universities graduate high-credential workers every year, but the economy does not produce a matching supply of high-credential jobs. "Graduate working minimum wage" is not individual misfortune. It is the system's inevitable output.
The logic of Korean parental investment in education was once clear. A good university, a good job, a stable middle-class life. Exam preparation of three to four years — extended further with repeat attempts. Four years of university, two to three years of job preparation — a total of nine to eleven years of time, plus private tutoring, university tuition, and job-preparation costs: an investment of ₩100 million to ₩300 million, roughly $70,000 to $220,000. At the end of that investment, the occupations entered are now exposed to the possibility of AI replacement within five to ten years. The OECD estimates that approximately 50 percent of Korean jobs are exposed to AI — with white-collar office work, sales, machine-assembly operations, and routine manual labor among the highest-risk categories. These are precisely the occupations into which Korean university graduates most commonly enter.
There is one further layer of irony. The government and the private tutoring market are investing in coding education with the message "prepare for the AI era." Korea's smart education market stands at $9.76 billion — approximately ₩13 trillion. The government is deploying more than ₩1 trillion in digital education infrastructure through 2025. But AI is developing at precisely the speed required to replace coding work. Boilerplate code writing, simple module development, and basic bug fixing — the tasks performed by junior developers — are already being automated. Households that spent hundreds of thousands of won on the message "learning to code guarantees your future" will witness that future being absorbed by AI.
In Chapter 5 we saw Israel's Unit 8200 teaching real-world AI capabilities to 18-year-olds. Israeli parents ask their children: "What question did you ask your teacher today?" Korean parents ask: "What score did you get?" Where Israel chose operational capability and Singapore chose systems design, Korea chose the securing of rank within a hierarchy. The target is admission to SKY — Seoul National University, Korea University, Yonsei University. The exits are the major chaebol's open recruitment process or the civil service examination. Both pathways include precisely the occupations AI will automate first. In the AI era, securing rank within a hierarchy is the fastest-depreciating asset.
6. The Same Era, Different Readings
The displaced are not the only ones.
Park Ji-hoon, 36, is CEO of the AI startup SecondBrain. A graduate of KAIST's School of Computing, he spent three years at Naver's AI research lab before founding his company. What he read was not simply a market opportunity. In 2021, working with GPT-3's API at Stanford, he identified something: large language models are strong on what is publicly available on the internet, but useless on any given organization's internal knowledge. Smarter than Google Search, but less capable than the company's own internal database.
American AI is English-centric. Chinese AI is Chinese-centric. Systems that process the internal knowledge of companies operating in Korean, Japanese, and Indonesian are a low priority for either. That linguistic and cultural gap is his space. As of 2024: 420 client companies, ARR of ₩3.8 billion — approximately $2.76 million. Korean clients account for 71 percent of the total, but 80 percent of revenue growth comes from Japan and Southeast Asia. What he watches most carefully is "dependence on domestic large-company contracts." Supplying Samsung or SK is attractive at first — but the moment one becomes a large-company subcontractor, the momentum for independent global expansion is lost.
The irony is that what Park Ji-hoon builds is a tool for structuring the tacit knowledge of people like Lee Jung-hoon. When a Japanese mid-sized manufacturer's employee said "our company's 20 years of technical know-how is now searchable," Park felt that what he wanted to make was becoming real. If Lee Jung-hoon's 28 years of experience had been stored in a system like SecondBrain, AI might not have replaced him. Or it might have learned his experience and replaced him far more completely. Who manages that boundary — no one has decided.
Choi Young-seok, 49, is CEO of battery materials company EC Materials. What he read was geopolitical signal. Eighteen months before the Inflation Reduction Act — IRA — passed the U.S. Congress, he had already begun redesigning his supply chain to reduce dependence on China. Cutting Chinese precursor dependency from 68 percent to 22 percent took two years. In Chapter 11 we saw Korea's battery companies caught by the IRA's FEOC provisions. Choi moved before those provisions arrived. He did not avoid risk. He read the direction of risk first.
Lee Jung-hoon and Park Ji-hoon live in the same country, the same era. Kim Su-jin and Choi Young-seok operate within the same economy. What is the difference? The displaced accumulated experience inside the existing structure. The Discerning detected the direction in which the structure itself was changing. What Lee Jung-hoon built over 28 years was mastery within the current system. What Park Ji-hoon read was the direction in which that system was turning.
Yet this must not be reduced to a difference in individual ability.
Baguus Sutomo, 38, drove a 150-ton dump truck at the nickel mine in Sorowako, Sulawesi, Indonesia. Monthly earnings: approximately 12 million rupiah — roughly ₩750,000. Three times the income of his father farming in Solo. He paid his two children's school fees and bought a small house in Sorowako. "As long as the mine is here, I'm fine," he thought. When Komatsu's autonomous vehicle system was introduced, 28 of the 70 members of the haulage team received notices of contract termination. Baguus was reassigned to remote monitoring support, and his salary was cut 30 percent. What he cannot understand is why he became poorer when Indonesia banned raw nickel ore exports and declared it would process the ore domestically. The processing plants were built by Chinese companies; the skilled technicians came from China. The government's announcement that the value-added from resources would rise had not reached Sorowako.
Lee Jung-hoon, Kim Su-jin, Baguus. Three people who do not know each other. A Korean manufacturing department head, a Korean bank deputy director, an Indonesian mine driver. But all three displacements rest on the same formula.
In Chapter 12 we saw Korea's demographic cliff and regulatory rigidity threatening the sustainability of indispensability. The stories of the displaced are the human face of that threat. The structure runs as follows. A manufacturing middle manager takes early retirement. Uses severance to start a self-employed business. Within two to three years, closes, trapped inside a revenue structure dependent on platforms. While assets are consumed, private tutoring costs for the children continue. The highly credentialed child graduates and joins a bank — but AI begins to replace that job category. One generation's displacement produces the next generation's displacement. This is Korea's chain of displacement.
Applying the core formula: the displaced are the concrete face of the formula's third stage — social instability. Technological innovation in the form of AI, passing through the concentration of capital in the form of the chaebol and the platforms, arrives at Lee Jung-hoon and Kim Su-jin and Baguus. The formula's fourth stage — institutional redesign — has not yet come. Just as Britain's Factory Acts arrived 70 years after the start of the Industrial Revolution, Korea's version of the Factory Act is adrift. The debate over a delivery platform fee ceiling, the delayed passage of the AI Basic Act through the National Assembly, the tightening of regulations on bank branch closures — all are fragmentary, all are symptomatic, all are late.
The Discerning are those who know the formula. The displaced are those who, standing in the middle of the formula's operation, do not know it is a formula. Attributing that difference to individual responsibility is the equivalent of holding the Lancashire handloom weavers accountable for failing to invent the power loom. When structure changes, redesigning the structure is the state's task. When the state fails at institutional redesign, the displaced keep being displaced.
Lee Jung-hoon's chicken shop opened again today.
Part III stops here. The technology of position (Chapter 11), the fracture of institutions (Chapter 12), the paradox of chaebol and soft power (Chapter 13), the face of the displaced (Chapter 14) — the diagnosis is complete. Now comes the time for design. Part IV examines what Korea must change to stand as an indispensable presence in the between — the prescriptions for that question begin in Chapter 15, The Great Transition.