Employment among South Koreans aged 15 to 29 fell by 285,000 from June 2022 to June 2026. Industries in the top half of the Bank of Korea’s AI-exposure ranking accounted for 268,000 of that decline, or 94 percent. Employment among workers in their 50s rose by 173,000 in the same industries (Bank of Korea).
Those numbers do not show that AI eliminated 268,000 jobs. They show something more specific. Employment weakened at the first rung of the career ladder in industries where generative AI can perform more tasks, while it rose for workers who already had experience.
A second Bank study supplies the other half of the mismatch. Korean companies say they plan to hire more AI workers, yet struggle to find experienced people. Korea appears to be narrowing the route that produces the workers its employers already say are scarce.
The age contrast survives a harder test
The headline count comes from National Pension subscriber records covering about 16 million workplace-based subscribers. It is an observed change, not the output of a population adjustment or a causal model. The Bank mapped occupation-level task exposure to 75 industries, ranked the industries into quartiles, then treated the top two quartiles as highly exposed (Bank of Korea).
The age groups are directly comparable by period, data source and exposure bucket. Their starting weights were not the same. High-exposure industries held 62.4 percent of youth employment in 2022, against 50.5 percent of employment among people in their 50s. Yet those industries produced 94 percent of the youth decline and 75.2 percent of the older group’s 230,000 gain (Bank of Korea). Exposure alone does not explain the difference in direction.
The Bank also ran a triple-difference regression with industry and monthly fixed effects. After November 2022, a 25-percentile increase in industry AI exposure was associated with 7.0 percent higher employment among workers in their 50s. Youth employment was 7.6 percentage points weaker under the same comparison. The result supports a seniority tilt after common industry and time effects are removed. It still does not identify AI adoption as the cause (Bank of Korea).
That distinction matters because the 268,000 figure and the adjusted regression answer different questions. The first measures where the losses occurred. The second tests whether the age gap moves systematically with exposure. Neither counts jobs destroyed by a model.
Automation removes practice as well as work
The mode of AI use sharpens the pattern. The Bank classified tasks using Anthropic research on Claude usage. Automation means delegating task execution to the system. Augmentation means a person keeps control while using AI to learn, improve a draft or verify work.
Youth employment declined most in industries with the highest automation share. No equivalent pattern appeared for augmentation. Industries in the top augmentation quartile had a milder youth decline than the other groups (Bank of Korea).
This is suggestive, not Korean usage telemetry. The classification imports task patterns from Claude records and applies them to Korean industries. It does not measure which Korean employer deployed which tool.
The mechanism is still coherent. Junior workers learn through research, drafting, basic analysis and simple coding. Automating those tasks saves money now but also removes supervised practice. Experienced workers retain the judgment, client knowledge and institutional context needed to check AI output. A company can rationally cut entry hiring and still bid for senior talent. When many companies do it, the saving becomes a labor-pipeline shortage.
Demand is waiting at the wrong end of the ladder
The Bank’s workforce study used Revelio data built from LinkedIn profiles. It covered about 1.1 million people with Korean work experience and more than 10 million job histories, defining an AI worker as someone listing at least one of 12 skills. It estimated 57,000 AI workers employed in Korea in 2024 (Bank of Korea).
The same study estimated that about 11,000 Korean AI workers were employed abroad, equal to roughly 16 percent of the relevant Korean AI workforce. That is a separate location estimate, not part of the domestic 57,000 count. Its linked-profile method will miss workers who do not maintain public professional profiles, so both figures are estimates rather than a census (Bank of Korea).
Compensation adds another constraint. The estimated domestic wage premium for possessing AI skills reached about 6 percent in 2024, compared with roughly 25 percent in the United States under the same model specification. The comparison controls for firm and year effects, but LinkedIn-derived wages and self-declared skills are imperfect measures. It still suggests the market rewards AI skills much more heavily abroad (Bank of Korea).
Employer demand is not theoretical. In an October 2025 survey of 400 Korean firms, 69.0 percent of large companies and 68.7 percent of midsize companies planned to expand AI hiring. Large firms named skilled-talent shortages as their leading recruitment obstacle, at 27.4 percent, followed by high salary expectations at 25.3 percent (Bank of Korea). The shortage sits among workers ready to contribute immediately, not among generic course graduates.
Rebuild the first rung
Demographics explain part of falling youth inflows. They cannot explain rising exits. Average monthly outflows from high-exposure industries increased from 3,700 before the pandemic to 4,900 in the four years through June 2026. The Bank also identifies pandemic-era over-hiring, sector normalization, remote work and weaker in-house training as competing explanations. The youth share of new hires was already falling before ChatGPT (Bank of Korea).
AI may be accelerating that older shift rather than creating it. Another Bank survey found time savings from generative AI were larger for less-experienced workers, evidence that augmentation can compress an experience gap instead of widening it (Bank of Korea).
The policy problem is therefore not preserving every entry-level task. It is paying firms to preserve learning while the task mix changes. Apprenticeships, shared training for smaller companies and support for senior mentoring would attach skill formation to real work. Short courses cannot create firm knowledge or judgment in exceptional cases.
Korea’s two shortages are connected. Employers lack experienced AI workers, while young people are losing access to the work that creates experience. Automation can improve the quarterly cost line. It cannot hire the senior engineer the market expects to exist five years later.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.