# From Resting Youth to Entry Bottlenecks: Forecasting Generative AI's Effects on Korea's School-to-Work Transitions

## Abstract

Generative artificial intelligence (AI) is changing not only existing jobs but also the entry routes through which young people acquire first work experience. This paper examines that pre-employment margin through Korea's "resting" youth, an economically inactive group whose main activity is recorded as resting rather than employment, job search, education, care, or illness. Using ten waves of Statistics Korea's May youth supplementary labor-force microdata for 2016-2025, the study estimates weighted profiles of youth aged 15-29 and develops a transition-bottleneck foresight framework linking inactivity to AI-enabled task restructuring. Resting youth increased from 260.6 thousand in 2016 to 395.7 thousand in 2025, even as the youth population fell from 9.38 million to 7.97 million. The rise is concentrated in the 20s, especially ages 22 and 25, and among school leavers, graduates, dropouts, and youth with disrupted first-job experiences. In 2025, 81.9% of resting youth had left school, and 72.6% had previous work experience. First-job non-entry and first-job tenure below one year are strongly associated with current resting status. These findings suggest that AI's future impact on resting youth is unlikely to operate mainly through direct technological unemployment. The more plausible mechanism is entry-task erosion: firms automate or compress junior tasks that created learning-by-doing. The paper contributes a transition-side approach to technological forecasting and argues for policies that redesign entry-level learning, AI-complementary career bridges, and early intervention.

Keywords: generative artificial intelligence; youth inactivity; school-to-work transition; NEET; entry-level jobs; technological forecasting; South Korea

## 1. Introduction

Generative artificial intelligence has shifted the technology and work debate from a narrow concern with routine automation to a broader concern with knowledge work, entry-level cognitive tasks, and the institutional design of labor-market transitions. Large language models can draft documents, summarize information, generate code, classify text, answer customer queries, and assist with data work. These functions overlap with tasks historically performed by junior workers, interns, clerical staff, and early-career professionals. The central question for youth labor markets is therefore not only whether AI replaces current workers. It is also whether AI changes the first tasks through which young people learn to become workers.

This paper studies that question through Korea's "resting" youth. In Korean labor-force statistics, "resting" refers to economically inactive individuals whose main activity is recorded as resting. The category excludes employment, unemployment, formal education, household care, illness, and several other activities. It is narrower than broad NEET status, but it is analytically valuable because it captures young people who are outside work and active job search without being in a standard human-capital investment category. The category has become a policy concern in Korea because it has grown even while the youth population has declined.

Korea is a useful case for technological forecasting and social change. It combines high educational attainment, rapid digital adoption, compressed competition for stable jobs, a large small-firm sector, and increasing concern over delayed school-to-work transition. The Bank of Korea has recently argued that the growth of resting youth cannot be reduced to an "overly high expectations" story. Educational gradient, career adaptability, nonemployment duration, and structural labor-market change are central. Separately, the Bank of Korea's work on AI and youth employment reports that recent youth job losses have been concentrated in AI-exposed industries. OECD and ILO studies similarly emphasize that AI exposure is mediated by task composition, complementarity, and institutional context rather than by whole-job replacement alone.

This paper asks four research questions.

1. How has the size and composition of Korea's resting youth changed between 2016 and 2025?
2. At which ages and transition points does resting status concentrate?
3. How do prior work experience, first-job timing, and first-job tenure relate to resting status?
4. What future risks does generative AI create for these transition points?

The empirical contribution is a microdata-based decomposition of resting youth using ten waves of May youth supplementary labor-force microdata. The theoretical contribution is an entry-bottleneck foresight framework. Instead of treating AI exposure only as a property of incumbent occupations, the framework examines how AI may affect the routes into work: the availability of junior tasks, the skill threshold for first employment, and the re-entry paths of youth who have already left a first job.

The findings show that resting youth are not a homogeneous group. In 2025, the largest policy-relevant segments are first-job non-entrants, early first-job leavers, re-entry mismatch cases, and long-duration low-activity youth. This segmentation matters for AI futures. Clerical and basic cognitive tasks may be directly automated; professional entry routes may remain attractive but require AI-complementary capability; service and manual routes may be less exposed to generative AI but still suffer from weak job quality and limited progression. A foresight approach must therefore distinguish technological exposure from transition vulnerability.

## 2. Literature and Conceptual Framework

### 2.1 From automation risk to task transformation

The labor-market literature has long distinguished between tasks and jobs. Autor, Levy, and Murnane (2003) argued that computerization substitutes for routine tasks while complementing nonroutine analytic and interactive work. Frey and Osborne (2017) extended automation-risk analysis to occupations, while Arntz, Gregory, and Zierahn (2016) emphasized that occupation-level measures can overstate displacement because many occupations bundle automatable and non-automatable tasks.

Generative AI deepens this task-based view. Language models and AI-enabled software can affect text, code, customer interaction, information search, summarization, classification, and design support. Eloundou et al. (2023) estimate broad task exposure to large language models in the United States. Felten, Raj, and Seamans (2021) provide an AI occupational exposure index showing that AI exposure is not limited to low-skill work. The ILO's work on generative AI similarly finds that clerical support work is especially exposed, while many occupations are more likely to be augmented than fully automated. The World Economic Forum (2025) forecasts large structural churn in jobs and skills by 2030, with simultaneous creation and displacement.

This literature suggests a key distinction for youth: high AI exposure does not necessarily mean job disappearance, but it can mean that entry-level task bundles change. If tasks that once trained novices are automated, the occupation may remain important while the route into it narrows.

### 2.2 Youth inactivity and the school-to-work transition

Youth inactivity has usually been studied through unemployment, NEET status, discouragement, skill mismatch, family resources, health, and local labor demand. Korea's resting category is a more specific status within economic inactivity. It does not capture all disadvantaged youth, but it identifies a group outside both work and active job search. That makes it valuable for understanding pre-employment and re-entry risk.

The school-to-work transition literature emphasizes first jobs, early tenure, and scarring. A delayed first job can reduce human-capital accumulation and weaken employer signals. A poor first job or early exit can damage confidence and make future matching more difficult. Repeated short jobs may also reduce career identity. These mechanisms matter for resting youth because the microdata contain school-leaving dates, first-job dates, first-job separation, prior job experience, and current activity status.

### 2.3 Entry-task erosion as a foresight mechanism

This paper proposes "entry-task erosion" as a mechanism linking generative AI to resting youth. Entry-task erosion occurs when organizations automate, compress, or outsource tasks that previously allowed young workers to enter an occupation and learn by doing. Examples include basic document drafting, routine customer response, data cleaning, research summaries, simple coding, bookkeeping support, scheduling, and clerical record maintenance.

The mechanism has three features. First, it can affect youth before they become employees. If entry routes narrow, some youth may never appear as displaced workers. Second, it can coexist with stable or growing demand for experienced workers. AI may complement senior workers while reducing the need for junior support tasks. Third, it makes transition institutions more important. Training, internships, apprenticeships, and supervised projects become substitutes for learning that previously happened inside firms.

This conceptual framework leads to a transition-side forecast: AI will increase risk for resting youth when their re-entry depends on entry-level tasks that are automatable, when their first work experience was short or low quality, and when they lack institutions that translate AI tools into occupational capability.

## 3. Data and Methods

### 3.1 Data source and population

The analysis uses ten annual CSV files from Statistics Korea's May youth supplementary labor-force survey for 2016-2025. The population is individuals aged 15-29. The survey weight is used for all estimates. Weighted counts are reported in thousand persons. The analysis is repeated cross-sectional; it does not track the same individuals over time.

Resting youth are identified by the variable for main activity status in the previous week, using code 11 for resting. The analysis also uses sex, exact age, education, current school status, economic activity status, previous-job experience, previous occupation, previous employment status, reason for leaving the previous job, school-leaving month, first-job month, first-job separation month, number of jobs after school leaving, employment desire, non-search reasons, and youth supplementary nonemployment activity.

### 3.2 Measures

The main outcome is an indicator equal to 1 when a respondent is resting. The denominator changes depending on the analysis. Some tables use all youth aged 15-29. Others use school leavers, economically inactive youth, or resting youth only.

Several transition variables are constructed from year-month fields. Months since school leaving equals survey month minus the most recent graduation, completion, or dropout month. Months to first job equals first-job month minus school-leaving month. First-job tenure equals first-job separation month minus first-job month. Implausible negative values are set to missing.

For interpretability, transition variables are grouped into bins: school-leaving elapsed time, first-job waiting time, and first-job tenure. The analysis also identifies high-risk cells by crossing age group, education, and school status.

### 3.3 Empirical strategy

The paper reports weighted descriptive statistics, transition decompositions, and a weighted linear probability model. The linear probability model is not a causal model. It is used only to summarize which variables remain associated with resting status after accounting for year, sex, age group, education, school status, region, and prior work experience in some specifications.

The foresight component links these microdata results to evidence on generative AI exposure and labor-market transformation. It does not claim that AI caused the 2016-2025 rise in resting youth. Instead, it asks which observed transition bottlenecks are likely to become more severe, less severe, or qualitatively different under AI diffusion.

## 4. Results

### 4.1 Resting youth grew despite youth population decline

Table 1 shows the central time trend. Korea's youth population aged 15-29 declined from 9.38 million in 2016 to 7.97 million in 2025. Over the same period, resting youth rose from 260.6 thousand to 395.7 thousand. The resting share of the youth population increased from 2.8% to 5.0%, and the resting share of economically inactive youth increased from 5.2% to 9.8%.

The trend is not a simple reflection of weak youth employment. Youth employment rates improved between 2016 and 2022 and remained above 2016 levels in 2025. The rise in resting status therefore reflects polarization within youth transitions: more youth are employed, but a larger inactive subgroup is disconnected from job search and formal human-capital investment.

**Table 1. Youth labor-market status and resting status, 2016-2025**

| Year | Youth population (000) | Employed (%) | Inactive (%) | Resting youth (000) | Resting share of youth (%) | Resting share of inactive (%) |
|---:|---:|---:|---:|---:|---:|---:|
| 2016 | 9,378.4 | 42.1 | 53.4 | 260.6 | 2.8 | 5.2 |
| 2017 | 9,301.9 | 43.0 | 52.6 | 285.8 | 3.1 | 5.8 |
| 2018 | 9,157.1 | 42.7 | 52.3 | 296.0 | 3.2 | 6.2 |
| 2019 | 9,072.8 | 43.6 | 51.6 | 348.5 | 3.8 | 7.4 |
| 2020 | 8,934.2 | 42.2 | 53.0 | 462.1 | 5.2 | 9.8 |
| 2021 | 8,798.5 | 44.4 | 51.0 | 392.1 | 4.5 | 8.7 |
| 2022 | 8,594.5 | 47.8 | 48.5 | 345.6 | 4.0 | 8.3 |
| 2023 | 8,415.9 | 47.6 | 49.5 | 385.7 | 4.6 | 9.3 |
| 2024 | 8,173.4 | 46.9 | 49.7 | 398.4 | 4.9 | 9.8 |
| 2025 | 7,973.6 | 46.2 | 50.5 | 395.7 | 5.0 | 9.8 |

### 4.2 The increase is concentrated in the 20s, especially around ages 22 and 25

Resting status is not evenly distributed across youth ages. In 2025, the resting rate was 0.8% among ages 15-19, 6.7% among ages 20-24, and 6.5% among ages 25-29. Exact-age profiles are more revealing. Resting rates increased sharply at age 22, from 4.1% in 2016 to 9.3% in 2025, and at age 25, from 3.2% to 9.4%. These ages align with school completion, interruption, military-service timing for men, first-job entry, and early re-entry after leaving a first job.

The decomposition by sex and age shows that the 2016-2025 increase came mainly from ages 25-29. Male 25-29 resting youth increased by 58.3 thousand and female 25-29 resting youth by 59.6 thousand. The contribution of ages 15-19 was negative because the youth population in that age band declined and resting remained rare.

**Table 2. Exact-age resting rates, selected ages**

| Age | 2016 rate (%) | 2025 rate (%) | Change (percentage points) | 2025 resting youth (000) |
|---:|---:|---:|---:|---:|
| 20 | 4.7 | 7.2 | 2.4 | 23.9 |
| 22 | 4.1 | 9.3 | 5.2 | 38.5 |
| 24 | 4.2 | 5.7 | 1.5 | 35.8 |
| 25 | 3.2 | 9.4 | 6.2 | 65.1 |
| 26 | 2.7 | 6.0 | 3.3 | 37.7 |
| 28 | 3.0 | 6.1 | 3.1 | 42.4 |
| 29 | 3.1 | 6.2 | 3.1 | 43.4 |

### 4.3 Resting status is a school-exit problem more than a student problem

In 2025, 81.9% of resting youth had graduated, dropped out, or completed school. Current students account for almost none of resting youth. The highest rates are found among young people on leave from school or after dropout. High-school graduates on leave aged 20-24 had a resting rate of 23.9%; high-school dropouts aged 20-24 had a rate of 19.8%; and high-school graduates on leave aged 25-29 had a rate of 17.6%.

This result matters because it changes the policy target. Resting youth are not mainly students temporarily taking leisure. They are disproportionately outside school or in weakly attached school statuses such as leave and dropout. Formal education systems, public employment services, and youth welfare programs therefore need stronger handoff mechanisms.

**Table 3. High-risk cells by age, education, and school status, 2025**

| Segment | Youth population (000) | Resting youth (000) | Resting rate (%) |
|---|---:|---:|---:|
| 20-24 / high school / leave | 239.3 | 57.3 | 23.9 |
| 20-24 / high school / dropout | 81.9 | 16.2 | 19.8 |
| 25-29 / high school / leave | 64.0 | 11.2 | 17.6 |
| 20-24 / high school / graduate | 333.0 | 41.5 | 12.5 |
| 25-29 / high school / dropout | 214.6 | 20.5 | 9.5 |
| 25-29 / junior college / graduate | 617.0 | 57.1 | 9.3 |
| 20-24 / university / graduate | 284.1 | 23.3 | 8.2 |
| 25-29 / high school / graduate | 565.5 | 43.0 | 7.6 |
| 25-29 / university / graduate | 1,526.9 | 85.0 | 5.6 |

### 4.4 First-job non-entry and short first-job tenure are key bottlenecks

Among school leavers in 2025, the estimated resting rate was highest among those with no post-school employment experience, at 19.1%. This group accounts for about one third of resting youth among school leavers. Youth with one prior job had a lower resting rate, 4.9%, while those with two or more jobs had rates between 5.5% and 7.6%. This indicates that the first employment match is a major threshold.

First-job timing also matters. Youth whose first job came after more than five years had a resting rate of 15.0%, compared with around 9-12% for shorter waiting periods. First-job tenure shows a clearer pattern. Those whose first job lasted 7-12 months had a resting rate of 12.0%, and those whose first job lasted 0-6 months had a rate of 9.9%. The rate falls after one or two years of tenure. Short first-job duration therefore appears to be a transition risk signal.

**Table 4. School-to-work bottlenecks and resting status, 2025**

| Transition indicator | Category | Population (000) | Resting rate (%) |
|---|---|---:|---:|
| Jobs after school leaving | No job experience, inferred | 567.5 | 19.1 |
| Jobs after school leaving | One job | 1,559.0 | 4.9 |
| Jobs after school leaving | Two jobs | 979.6 | 7.6 |
| Jobs after school leaving | Four or more jobs | 558.5 | 6.5 |
| First-job waiting time | 0-3 months | 935.1 | 9.2 |
| First-job waiting time | 2-5 years | 270.3 | 11.6 |
| First-job waiting time | More than 5 years | 37.6 | 15.0 |
| First-job tenure | 0-6 months | 835.4 | 9.9 |
| First-job tenure | 7-12 months | 633.2 | 12.0 |
| First-job tenure | 1-2 years | 465.9 | 7.1 |
| First-job tenure | 2-3 years | 201.8 | 5.9 |

### 4.5 Resting youth are often re-entry cases, not only never-employed youth

In 2025, 72.6% of resting youth had previous work experience. Among resting youth with an identifiable previous occupation, the largest prior occupational groups were clerical support, service, professional, elementary, and plant-machine operator jobs. Previous employment status was split between temporary workers and regular employees, with temporary workers at 43.6% and regular employees at 43.3%. Among cases with a recorded reason for leaving, poor working conditions accounted for 37.9%, personal or family reasons for 37.1%, and completion of temporary or seasonal work for 16.4%.

These patterns indicate that resting status is not simply a pre-work condition. It is also a re-entry condition after low-quality, short, or poorly matched jobs. This is important for AI forecasting because AI may not only block first entry; it may also change the quality and learning content of early jobs.

**Table 5. Prior work characteristics among resting youth, 2025**

| Characteristic | Category | Share among valid cases (%) |
|---|---|---:|
| Previous occupation | Clerical support | 27.8 |
| Previous occupation | Service | 26.1 |
| Previous occupation | Professional | 14.7 |
| Previous occupation | Elementary | 11.9 |
| Previous occupation | Plant-machine operator | 8.4 |
| Previous employment status | Temporary employee | 43.6 |
| Previous employment status | Regular employee | 43.3 |
| Previous employment status | Daily worker | 8.1 |
| Reason for leaving | Poor working conditions | 37.9 |
| Reason for leaving | Personal or family reasons | 37.1 |
| Reason for leaving | Temporary or seasonal work ended | 16.4 |

### 4.6 Resting youth contain distinct policy types

The internal composition of resting youth is heterogeneous. In 2025, only 16.5% expressed a desire for employment, but this group was older, more work-experienced, and more likely to be in the 25-29 age group. Resting youth with nonemployment duration of three years or more were more male, more concentrated among high-school graduates or less, and more likely to report simply spending time. Youth reporting job education, exam preparation, or job search within the youth supplement had much higher tertiary-education shares.

This supports a four-type interpretation. First, first-job non-entrants are youth who have left school but have not crossed the first employment threshold. Second, early-exit youth left an initial job quickly and may carry negative signals or weakened confidence. Third, re-entry mismatch youth have work experience and some employment desire but face condition, location, skill, or job-search barriers. Fourth, long-duration low-activity youth have weak labor-market attachment and require more than job matching.

**Table 6. Internal profiles of resting youth, 2025**

| Resting-youth group | Size (000) | Age 25-29 (%) | High school or less (%) | Prior work (%) | Simply spending time (%) |
|---|---:|---:|---:|---:|---:|
| All resting youth | 395.7 | 55.4 | 55.1 | 72.6 | 47.9 |
| Want employment | 65.2 | 72.4 | 48.8 | 87.6 | 48.8 |
| Do not want employment | 318.8 | 51.3 | 56.1 | 69.5 | 47.4 |
| Nonemployed less than 1 year | 170.5 | 64.2 | 39.4 | 89.0 | 55.6 |
| Nonemployed 3 years or more | 59.5 | 74.5 | 68.1 | 51.2 | 69.3 |
| Job education, exam, or search | 69.7 | 64.3 | 23.8 | 80.9 | 0.0 |

## 5. Foresight Analysis: AI and the Future of Resting Youth

### 5.1 Why AI risk appears before observed displacement

Standard automation studies usually assign exposure to jobs currently held by workers. Resting youth require a different approach because they are not currently matched to jobs. Their risk lies in prospective entry routes. If entry-level tasks are automated, young people may not be fired; they may simply not be hired, or they may stop searching because accessible jobs no longer create credible progression.

The microdata identify where this risk is likely to appear. It appears at school exit, in first-job non-entry, in very short first-job tenure, and in re-entry after low-quality or temporary work. These are precisely the places where organizations decide whether to train novices or use technology to substitute for novice tasks.

### 5.2 Three AI futures for resting youth

Scenario 1 is the augmented-bridge scenario. Firms and public employment services use generative AI to reduce training costs and create structured entry routes. Resting youth learn document automation, workflow verification, customer-response quality control, data cleaning, and domain-specific AI use through supervised projects. In this scenario, AI lowers the cost of re-entry.

Scenario 2 is the entry-erosion scenario. Firms automate junior clerical, reporting, customer-response, and simple analytical tasks without building new learning routes. Experienced workers become more productive, while fewer juniors are hired. Resting youth with high-school, junior-college, or general university backgrounds face a higher threshold for first employment. In this scenario, resting status becomes more persistent even if aggregate productivity improves.

Scenario 3 is the low-quality fallback scenario. Youth avoid AI-exposed clerical and professional pathways and move toward service, logistics, personal care, or manual jobs. These jobs may be less directly exposed to generative AI, but many offer weak wages, irregular hours, and limited career ladders. In this scenario, the resting rate may stabilize, but transition quality and long-term earnings deteriorate.

The available evidence suggests that Korea faces all three possibilities. OECD/Korea Labor Institute analysis finds that Korea's AI adoption remains uneven but that task exposure and skill demand are increasing. The ILO emphasizes clerical exposure. WEF forecasts large skill disruption. The Bank of Korea's AI-youth analysis points to early-career vulnerability in AI-exposed industries. The policy question is therefore not whether AI is good or bad for youth, but whether institutions convert AI into structured entry learning or allow it to remove entry tasks.

### 5.3 Occupation-specific implications from prior-job profiles

Previous occupations among resting youth provide clues about vulnerable pathways. Clerical support and service jobs are the two largest recorded previous occupational groups among resting youth. Clerical jobs are directly exposed to document automation, data entry automation, scheduling, basic accounting, record processing, and AI-assisted correspondence. Service jobs are more heterogeneous. Some routine customer-interaction jobs may be exposed to chatbots, kiosks, and automated ordering, while relational service and care work may remain human-intensive.

Professional prior jobs account for 14.7% of valid previous occupations among resting youth. For this group, the risk is not simple disappearance. It is rising entry standards. AI tools can increase productivity for workers who already possess domain knowledge, but they can also reduce the need for junior workers doing drafting, coding, research, or support tasks. Resting university graduates therefore need AI-complementary portfolios and supervised projects, not only job-search counseling.

Plant-machine operators and elementary workers face a different technology mix: robotics, digital monitoring, logistics automation, and cyclical demand rather than only generative AI. These groups require regional industrial policy and training tied to concrete employer demand.

## 6. Discussion

The findings reframe resting youth as a transition-system issue. The common interpretation of resting youth as passive or preference-driven misses the concentration at specific transition points: leaving school, failing to obtain a first job, leaving a first job within one year, and attempting to re-enter after low-quality employment. These are not merely individual states. They are institutional handoff failures.

This framing also changes the AI debate. If AI is assessed only through current employment, resting youth remain invisible. Yet they may be among the most affected because AI can alter the first tasks they need in order to become employable. This paper therefore contributes to technological forecasting by shifting the unit of analysis from incumbent job exposure to entry-route exposure.

The empirical results also caution against treating low AI exposure as safety. High-school graduates, dropouts, and some service or manual pathways may be less exposed to language-model substitution, but they still face weak job quality, lower career mobility, and higher risk of long-duration inactivity. Conversely, professional and clerical pathways are AI-exposed in different ways. Clerical work faces direct substitution pressure; professional work faces entry-threshold escalation.

Finally, the 2025 profile of resting youth suggests that policy must be segmented. A youth who has never obtained a first job, a youth who left a first job after six months, a youth who wants re-entry after previous work, and a youth who has been inactive for three years require different interventions. A generic activation program cannot address all of these mechanisms.

## 7. Policy Implications

First, employment services should triage resting youth by transition bottleneck rather than by age alone. The relevant categories are first-job non-entry, early first-job exit, re-entry mismatch, and long-duration low activity. Each category requires different timing, incentives, and support intensity.

Second, AI literacy should be occupationally embedded. For clerical pathways, training should focus on AI-assisted document workflows, verification, privacy, cybersecurity, and exception handling. For professional pathways, programs should require portfolio projects using domain-specific AI tools. For service pathways, training should focus on human-AI coordination in customer service, hospitality, care, logistics, and retail operations.

Third, policy should preserve entry-level learning tasks. If firms automate junior tasks, public policy can support structured apprenticeships, supervised project work, and wage subsidies conditional on training quality. The goal is not to protect every old task. The goal is to replace lost learning-by-doing with new learning institutions.

Fourth, early intervention should focus on high-risk cells: high-school graduates on leave, high-school dropouts, junior-college graduates, and 25-29-year-olds with prior job experience but weak re-entry. These groups appear repeatedly in the microdata.

Fifth, long-duration resting youth need integrated support. For those inactive for three years or more, job matching alone is unlikely to be sufficient. Counseling, mental-health support, social connection, basic skill rebuilding, and small-step work exposure are likely prerequisites for AI training or formal employment programs.

## 8. Limitations and Future Research

The study has several limitations. First, the data are repeated cross-sections, so individual transitions cannot be observed. Panel data would allow direct estimation of whether first-job waiting time, short tenure, or school dropout predicts later resting status for the same individual.

Second, the AI analysis is a foresight interpretation rather than a causal estimate. The study does not claim that AI caused the 2016-2025 increase in resting youth. It identifies transition points that are likely to be affected by AI diffusion.

Third, previous occupation is available only for those with recorded prior work and valid occupation codes. Broad occupational categories hide within-group variation. Future research should link detailed occupation codes to task-level AI exposure indices.

Fourth, the analysis does not include household income, mental health, local labor demand, or firm-level AI adoption. These factors are important for explaining resting status and should be integrated when data permit.

Fifth, the May survey captures a specific timing in the Korean labor-market calendar. Seasonal and institutional timing may matter, especially around graduation and recruitment cycles.

## 9. Conclusion

Generative AI changes the future of youth labor markets by changing the entry routes into work. Using ten waves of Korean May youth supplementary labor-force microdata, this paper shows that resting youth grew from 260.6 thousand in 2016 to 395.7 thousand in 2025 despite a shrinking youth population. The increase is concentrated in the 20s, especially around ages 22 and 25, and is strongly connected to school exit, first-job non-entry, short first-job tenure, prior low-quality jobs, and long-duration inactivity.

The central forecast is not mass technological unemployment among resting youth. The more plausible risk is entry-task erosion. AI can remove or compress the junior tasks that previously allowed young people to acquire experience, signals, and confidence. If institutions do not replace those learning routes, resting status may become more persistent. If institutions use AI to create structured bridges, however, the same technology can lower re-entry costs.

For technological forecasting and social change, the implication is clear: future-of-work analysis should not stop at incumbent worker exposure. It must also examine those outside employment whose pathways into work are being reshaped before they arrive.

## Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this draft, the author used OpenAI Codex/ChatGPT to assist with data processing, literature organization, and language drafting. The author must review and edit the content as needed and takes full responsibility for the content of the submitted manuscript.

## Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

## Declaration of competing interest

The author should complete Elsevier's declaration form before submission. No competing interest is declared in this draft.

## Data availability statement

The analysis uses Statistics Korea May youth supplementary labor-force microdata files for 2016-2025 supplied by the user. Redistribution of raw microdata should follow the conditions of the source data provider. Derived tables used in this manuscript are reported in the paper.

## References

Acemoglu, D., and Restrepo, P. (2018). Artificial intelligence, automation, and work. In A. Agrawal, J. Gans, and A. Goldfarb (Eds.), The Economics of Artificial Intelligence: An Agenda. University of Chicago Press.

Arntz, M., Gregory, T., and Zierahn, U. (2016). The risk of automation for jobs in OECD countries: A comparative analysis. OECD Social, Employment and Migration Working Papers, No. 189. OECD Publishing.

Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3-30.

Autor, D. H., Levy, F., and Murnane, R. J. (2003). The skill content of recent technological change: An empirical exploration. Quarterly Journal of Economics, 118(4), 1279-1333.

Bank of Korea. (2025). AI diffusion and youth employment contraction: Seniority-biased technological change. BOK Issue Note 2025-30.

Bank of Korea. (2026). Characteristics and assessment of resting youth: A comparative analysis by type of nonemployment. BOK Issue Note 2026-3.

Eloundou, T., Manning, S., Mishkin, P., and Rock, D. (2023). GPTs are GPTs: An early look at the labor market impact potential of large language models. arXiv preprint arXiv:2303.10130.

Felten, E., Raj, M., and Seamans, R. (2021). Occupational, industry, and geographic exposure to artificial intelligence: A novel dataset and its potential uses. Strategic Management Journal, 42(12), 2195-2217.

Frey, C. B., and Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254-280.

Gmyrek, P., Berg, J., and Bescond, D. (2023). Generative AI and jobs: A global analysis of potential effects on job quantity and quality. ILO Working Paper 96. International Labour Organization.

Gmyrek, P., Winkler, H., and Garganta, S. (2025). Generative AI and jobs: A refined global index of occupational exposure. International Labour Organization.

OECD and Korea Labor Institute. (2025). Artificial Intelligence and the Labour Market in Korea. OECD Publishing.

Statistics Korea. (2016-2025). Economically Active Population Survey, May Youth Supplementary Survey microdata. Statistics Korea MicroData Integrated Service.

Statistics Korea. (2025). Results of the May 2025 supplementary survey on youth in the Economically Active Population Survey. Statistics Korea.

World Economic Forum. (2025). The Future of Jobs Report 2025. World Economic Forum.