The Korean labour market after AI: job-postings analysis
AI 등장 이후 한국 고용 시장 (채용공고 분석)
A · Strong evidence Underscore with economists from Kyoto University and KDI analyse 10.4 million Korean job postings with within-firm fixed effects, placebo tests and a public working paper; strong method, but one platform and not yet peer reviewed.
Summary based on the full text·Original: about 25 minutes to read
In short
An interactive report by data journalism outfit Underscore with MBC's PD Note (2026), by TaeYoung Kang (Underscore), Hyunjae Kang (Kyoto University economics) and Sunham Kim (KDI), with a working paper "Raising the Bar" on SSRN. It covers 10.4 million postings from one of Korea's largest job platforms. Postings open to new graduates fell from 2022 across office and professional jobs, but that fall began before ChatGPT and overlaps rate rises. The core finding is different: after ChatGPT, the more of a job generative AI can do, the more experience employers began to demand, even comparing postings from the same firm in the same month.
Key numbers
- 10,413,271job postings analysed (July 2019 to January 2025) from 434,688 firms on one Korean platform
- 3.73 to 5.18 yearsaverage required experience in developer postings, summer 2022 versus January 2025
- +0.073 yearsextra rise in required experience per 1 SD of AI exposure after ChatGPT, same firm, month and job family
- -33.5%change in new-graduate developer postings after ChatGPT versus blue-collar control (PPML DiD)
- -0.77 ppdrop in share of postings open to new graduates per 1 SD of AI exposure
As reported by the source.
Korean·Web page·opens korean-labor-market-after-ai.vercel.app
What it reports
- Developer new-graduate postings doubled from a spring 2020 low to a summer 2022 peak, then fell from the second half of 2022, before ChatGPT's release at end-November, to below the earlier low by January 2025.
- Against a blue-collar control (PPML difference-in-differences), only developers (-33.5%) and digital design (-20.6%) showed significant falls in new-graduate postings; both pass a placebo test at a fake July 2021 break. Experienced-hire postings showed no significant change in any group.
- The fall hit tech firms (-48%) and non-tech firms (-24%), and was significant for SMEs (-35%) but not startups (-11%), against the "startups lost funding" explanation.
- Developer required experience held at about 3.7 years from 2019 to 2022, then rose to 4.07 in December 2022, 4.82 in 2024 and 5.18 in January 2025 (+0.66 years versus blue-collar).
- Main model (job, firm-by-month and job-family-by-month fixed effects): +0.073 years per 1 SD of AI exposure after ChatGPT (p<.01); +0.055 years excluding IT jobs and IT firms; placebo +0.009. The gap became significant from autumn 2023 and widened to +0.08 to +0.11 years in 2024.
- Per 1 SD exposure, the share of postings open to new graduates fell 0.77 points and required years within experienced postings rose 0.088.
- Controls for rate sensitivity, the 2021 hiring boom, prior skill levels (education, pay) and firm industry did not remove the effect; with skill controls it grew to +0.107 years.
- Employment insurance data: under-30 employment in information and communications peaked in the second half of 2022 and fell for three years while workers in their 30s and over 50 rose.
How they got it
15.1 million postings collected from February 2017 to April 2025; the 10.4 million posted July 2019 to January 2025 were analysed. AI exposure per job uses Eloundou et al. (2023) task scores linked through 1,920 job keywords to O*NET occupations (some matched with AI assistance and human review). Firm exposure was scored by GPT-5.6 Astra and Claude Opus 5 (correlation 0.983). Errors are two-way clustered by firm and job.
Limits
- One platform; firms using their own career pages or other boards are missing.
- Required experience in a posting is not the experience of the person hired.
- Older postings have higher deletion rates (about 30% for 2017-2018), hence the 2019 start.
- Keyword-to-occupation matching involves judgement; the authors report robustness to alternative aggregation.
- Working paper, not peer reviewed; a trend-break test at ChatGPT's launch was not significant, consistent with a gradual effect.
Why it matters for companies
This is one of the cleaner tests of the junior-displacement question outside the US, and its answer is narrower than headlines: employers raised experience bars for AI-exposed jobs rather than simply posting fewer of them. For workforce planners the risk is a thinning entry pipeline in software, design and office roles. Firms that stop hiring juniors now will face a shortage of mid-level staff in a few years unless they redesign entry roles and training.
Key passages in translation
Korean original
생성형 AI가 대신할 수 있는 일이 많은 직무일수록 ChatGPT 등장 이후 더 많은 기간의 경력을 요구하는 변화가 시작되었습니다.
The more of a job's work generative AI can take on, the more years of experience employers began to require after ChatGPT appeared.
Korean original
신입보다는 경력을, 같은 경력직 내에서도 더 높은 연차를 선호하게끔 채용 시장이 변화했다는 것입니다.
The hiring market shifted to prefer experienced hires over new graduates, and, among experienced hires, more senior ones.
Korean original
채용 공고와 고용이 완벽히 일치하는 개념은 아닙니다.
Job postings and employment are not exactly the same thing.
Summary based on the full text. Read the original for the full argument and data.Read the original on korean-labor-market-after-ai.vercel.app