Dispatches
Essays··8 min read

Seoul Pays for the Rung It Removes

South Korea funds AI adoption subsidies and youth employment recovery from the same treasury, without connecting the two. A Konkuk University forum paper finds voucher recipients cut new hiring of workers aged 19–34 by roughly 10%; Bank of Korea data show 94% of youth job losses in AI-exposed industries. The 2027 budget answers mostly with classroom training — what is missing is a reporting condition on the voucher and a first-job co-payment inside firms that bought the software.

What does a state owe a twenty-four-year-old whose first job it helped pay to automate?

I mean the question literally, as a line in a budget. South Korea now funds both sides of that transaction from one treasury, through two ministries that rarely share a press release. The Ministry of Science and ICT gives small firms money to buy AI. The Ministry of Employment and Labor gives young people money to recover from what those firms then stop doing, which is hiring them. Each programme makes sense on its own page. Put the pages side by side and it starts to resemble paying one contractor to remove a staircase and another to run climbing lessons.

Note on a forum paper, Seoul, 9 November

The clearest evidence arrived a week ago and has not been through peer review, so it deserves attention and caution in equal measure. Lee Hwan-woong, an economist at Konkuk University, presented an analysis of the science ministry's AI Voucher programme at a policy debate co-hosted by four Korean labour and law associations. As reported by the Seoul Economic Daily on 11 November, he compared 636 firms that won vouchers with 1,603 that applied and were rejected, using corporate panel data from 2015 to 2024. Recipients that adopted AI cut total new hiring by 8% relative to the rejected applicants. New hiring of workers aged 19 to 34 fell 10.1% over the three years after adoption.

For a small firm the voucher is serious money. In 2026 a demand company could receive up to ₩200 million toward AI solutions and consulting, under a scheme run with the National IT Industry Promotion Agency.

Three caveats before anyone builds on Lee's numbers. It is a forum presentation, not a refereed article. Comparing winners with losers of a selection process invites selection bias, since the agency presumably chose firms it judged likelier to succeed, and the press account does not say how he handled that. And the 10.1% applies to adopters, with no baseline headcount reported. I would want the full paper before quoting it to a cabinet. I would not wait for it before asking the question it raises.

Lee's proposed mechanism should worry a finance ministry more than a labour ministry. Korea makes dismissing regular staff hard, so a firm that installs software for document sorting, routine data processing or basic inspection dismisses no one. It leaves a retirement unfilled and trims next year's graduate intake. The adjustment shows up in the flow of hiring, which most employment impact assessments do not track, and stays out of the stock of employment, which they do.

Note on two Bank of Korea releases, August and October

The national data fit that mechanism uncomfortably well. In August the Bank of Korea's labour market research team reported that jobs held by 15-to-29-year-olds fell by 285,000 between June 2022 and June 2026, and 268,000 of those losses, 94%, were in industries it classes as highly exposed to AI. Employment of people in their fifties rose by 230,000 over the same window, three quarters of it in those same industries. When an earlier version of the study covered July 2022 to July 2025, the bank said Korea showed "the early signs of transforming into a seniority-biased labor market."

Oh Sam-il, who heads the BOK team, gave the reason plainly: AI "primarily performs tasks that are laid out in manuals, which means it can replace younger workers." Manual tasks are what juniors are hired to do while they learn the parts no manual covers.

The newest figures point the same way. On 11 November the Ministry of Data and Statistics reported that October employment rose by 139,000 year on year while employed 15-to-29-year-olds fell by 196,000, putting the youth employment rate at 42.9%. The Korea Herald adds that 46% of employed young people held nonregular contracts as of August, the highest share since the series began in 2003, and that 67.6% of 500 firms surveyed by the Korea Enterprises Federation named work experience as the first thing they look for in a recruit.

That last figure closes the loop. Firms want experience. The cheapest experience used to be the entry-level job. The entry-level job is what is being automated, partly with public money.

Note on the 2027 labour budget

Seoul has noticed. The labour ministry's 2027 proposal, ₩38.95 trillion in total and up 3.4%, lifts the youth employment budget from about ₩2.6 trillion to about ₩3.3 trillion, a 27% increase and the largest since 2021. K-Digital Training rises from ₩521.3 billion to ₩608.9 billion, with AI training places doubling from 10,000 to 20,000. The K-New Deal Academy grows to 50,000 participants. Korea Polytechnics gets ₩56 billion for a youth stepping-stone campus. In July a jobs task force co-chaired by the finance and labour vice ministers set a target of more than 200,000 specialists in advanced industries by 2030, and on 6 November the labour ministry outlined a new first-employment support system spanning career exploration, work experience, training and placement.

Inside that budget sits the line I think matters most, and it is the smallest: a new Youth AI First Job Career Support pilot, ₩14.5 billion, for 750 people.

₩14.5 billion for 750 first jobs. ₩608.9 billion for training places.

Divide it out and the pilot spends roughly ₩19 million per participant. That is expensive per head. It is also the only line in the package aimed squarely at the missing object, a first job inside a firm where someone is paid to learn on real work with AI alongside. Classroom places teach the tools. They do not put a graduate on a payroll, and the employers' own survey says the payroll history is what they screen for.

Note on the strongest objection

The best case against my reading comes from the Bank of Korea itself. In its 26 October report on youth job quality, covered by The Korea Times, the bank argued that the problem is structural and older than generative AI. Semiconductor-led growth produces output without many regular jobs, most regular hiring still happens in non-IT manufacturing and services, and Korea ranks tenth of 23 major economies on the OECD measure of protection against individual dismissal. Its verdict: "Measures aimed solely at improving labor supply would be insufficient to resolve the underlying problems." The August report likewise cautioned that weak youth hiring cannot be pinned on AI alone, citing post-pandemic industry shifts and a preference for experienced hires.

On the diagnosis the bank is right, and its warning about supply-side measures cuts against the training-heavy shape of the 2027 budget at least as much as against anything I am arguing. Where I part company is on what follows. Dismissal law in Korea is a long political negotiation with no assured outcome. Rigidity explains why the AI shock lands on new hires and spares incumbents; it gives no reason for the state to keep subsidising that shock without measuring it. The voucher's terms can change in one budget cycle. And if chip-led growth keeps producing output with thin regular hiring for years, the entry flow that remains becomes more valuable to protect.

Note on the stake

Here is the position I would defend. Korea's AI adoption subsidies should carry the employment-flow reporting Lee proposes: recipient firms report new hires by age band for three years after the voucher, separately from total headcount, with the data going to the labour ministry as a condition of payment. That costs almost nothing and turns one forum paper into a running measurement. Where a recipient's junior hiring falls, that firm should be first in line for the first-job pilot, with the state co-paying a trainee's wage inside the company that bought the software.

Over time I would also shift the proportions. At 750 places the first-job pilot is a test. If its participants are still employed two years on, it deserves a share of the money now going to classroom seats, because it buys what the market has stopped producing. If it fails, better to learn that at ₩14.5 billion than at ten times the sum.

Underneath, this is a question of relative prices, and Lee's sharpest point is about them. Hiring raises fixed costs such as social insurance contributions, while tax credits and vouchers reward capital and automation. A government that cuts the price of software and leaves the price of a twenty-four-year-old untouched should not be surprised by which one firms buy.

Note to file

Korea is paying twice for the same rung. One ministry subsidises small firms to adopt AI, and the first quasi-experimental look at that programme, from Konkuk University and still unpublished, finds recipients cutting hiring of 19-to-34-year-olds by about a tenth. Bank of Korea data show youth jobs falling almost entirely in AI-exposed industries while workers in their fifties gain there, and the central bank stresses that dismissal rules and chip-led growth matter as much as the technology. The labour ministry is responding with a youth budget 27% larger, mostly for training places, plus one small pilot that buys first jobs directly. Taken together, the record describes a hiring flow that the subsidy system pays into at one end and out of at the other, without yet counting either.


Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.

Cartouche
Seoul Pays for the Rung It Removes · Dispatches, 16 November 2026 · T. Singh