The AI Job Crisis Starts with Entry-Level Hiring, Not Layoffs

News about companies firing employees due to AI is rare, but entry-level positions are quietly disappearing. This suggests that the AI job shock is manifesting in entry-level hiring long before it reaches the stage of layoffs.

As of May 2026, the number of employment insurance subscribers under 30 in Korea dropped by 2.8% year-on-year, and U.S. payroll data shows a 13% relative decline in hiring for 22-25 year-olds in occupations with high AI exposure.

Why is the AI job shock hitting entry-level hiring first instead of layoffs?

Because freezing hiring is the quietest choice a company can make.

To lay off existing employees, a company must consult with labor unions, pay severance, and deal with the risk of negative press. For example, if a firm reduces its annual intake of new recruits from ten to six, no one outside the company notices. The rejected applicants simply remember that year’s competition as being particularly fierce.

Corporate surveys point in the same direction.

At a youth employment seminar held by the Federation of Korean Industries (FKI) on August 24, KDI researcher Han Yo-seop presented survey data showing that more companies expected to reduce new hiring due to AI adoption than expected to adjust existing staff levels (MTN Report).

While reading KERI Focus No. 2, which summarized the July 7 FKI seminar on “The Coexistence of AI and Jobs,” I paused at the section on entry-level jobs for youth. Professor Lee Chul-hee of Seoul National University pointed out that the core issue is the decline in entry-level positions for young people due to the spread of AI (Herald Business).

Layoffs make headlines, but those who are never hired leave no record behind. This is why it is easy to miss this shift if you only look at unemployment rate graphs.

Korean entry-level hiring data: A decline starting with under-30 employment insurance subscribers

In the May 2026 administrative statistics for employment insurance, the number of subscribers under 30 was approximately 2.233 million, a decrease of 65,000 (2.8%) from a year ago (Segye Ilbo). By industry, the information and communications sector saw a 9.3% decline, and professional, scientific, and technical services fell by 4.1%. Narrowing down to specific roles, 20-something software developers dropped by 10.1%, designers by 7.6%, and legal assistants under 30 fell for 13 consecutive months to 7,175.

However, these figures only show the number of subscribers, or how many jobs remain. They do not distinguish whether the decrease is due to fewer new hires or more departures, and the government has not definitively attributed the decline to AI.

Signals from hiring data tell a different story. In the same report, based on the second half of 2025 business surveys, job openings for computer system specialists dropped by 27.5%, designers by 15.9%, and writers/translators by 32.7%. Looking specifically at software developers, a Korea Labor Institute report shows that the proportion of entry-level job postings fell from 53.5% in 2022 to 37.4% in 2024 (News Tomato). It is notable that entry-level job postings have declined more sharply than those for experienced hires.

U.S. AI job data also shows a decline in 22-25 year-old entry-level positions first

If we only had Korean statistics, one might blame the economic downturn.

However, a “Canaries in the Coal Mine” paper (August 2025) by Professor Erik Brynjolfsson’s team at the Stanford Digital Economy Lab, which analyzed payroll records from a major U.S. payroll software company, presents nearly the same picture. After generative AI became widely used, employment of 22-25 year-olds in occupations with the highest AI exposure fell by a relative 13%, even after controlling for firm-level shocks. Employment of experienced workers in the same fields and youth in low-AI-exposure fields remained stable or continued to grow.

The decline was concentrated in occupations where AI is used to replace human work.

In occupations where AI is used to assist humans, this decline was not as pronounced. The paper is frequently cited for noting that, within specific roles, employment for 22-25 year-old software developers dropped by nearly 20% compared to its peak at the end of 2022. The title of the study is apt: just as canaries were the first to detect toxic gas in coal mines, entry-level positions serve as a precursor for the broader labor market.

Why do AI jobs bypass areas with labor shortages?

There is something strange here.

People often say that Korea needs AI because there aren’t enough people to work. If that were the case, AI should be filling the gaps where labor is most scarce.

In reality, it is the opposite.

According to Professor Lee Chul-hee, AI exposure is relatively higher in high-wage, high-skill sectors like finance and law. Sectors where labor shortages are expected due to demographic shifts—such as caregiving, transportation, and dining—are difficult to replace with AI. His diagnosis is that AI alone cannot solve this supply-demand imbalance. Instead, we are seeing automation in areas where there is an abundance of labor, while areas with labor shortages remain empty.

In fact, this mismatch has been a long-standing puzzle for AI researchers.

In the 1980s, roboticist Hans Moravec pointed out that while logical tasks like chess are easy for computers, tasks like walking, picking up objects, or recognizing surroundings—things even a one-year-old can do—are incredibly difficult. This is known as the Moravec’s Paradox.

The explanation is that abstract reasoning, which humans find difficult, is a relatively recent evolutionary trait and thus easier to transfer to machines, whereas physical tasks are skills honed by a very long history of evolution and are difficult to mimic.

A caregiver helping an elderly person turn over in bed or a delivery driver navigating a hallway without an elevator to drop off food cannot be solved by a language model’s writing prowess. Comparing contract clauses or summarizing financial statements is the exact opposite.

If the bottom rung of the career ladder is removed, what will happen to AI jobs in 10 years?

The contraction of entry-level hiring is frightening not because of the present, but because of the future.

Even the 30- or 40-something experienced workers currently reviewing AI drafts were once entry-level employees who wrote those drafts themselves. They gained judgment by correcting errors and rewriting the same reports over and over.

This is why tacit knowledge—which is difficult to record in documents—is often cited as the reason why experienced hires are still in demand.

A ladder with the bottom rung missing may still stand for a moment because people are still on the higher rungs.

The problem is what happens on the day those people retire. The generation that never stepped on the first rung will not be able to climb to the top.

Similar forecasts have emerged from the small and medium-sized enterprise (SME) sector.

Noh Min-sun, a researcher at the Korea Small Business Institute, noted that if AI is fully adopted by SMEs, there is a high possibility that demand for labor will shrink, leaving only a demand for highly specialized roles. The suggestion was made that since AI-based solo entrepreneurship might increase, support for youth to build AI-utilization skills is necessary. This means that while the number of businesses might grow, the number of companies willing to hire entry-level staff may decrease.

Most AI pilots have stalled, so why has entry-level hiring decreased?

There is a figure that seems contradictory at first glance.

According to an MIT NANDA Initiative report, 95% of corporate generative AI pilots failed to deliver results (Fortune Report, August 2025). This figure implies that only about 5% of pilots drove rapid revenue growth; translating this as “stalled” or “failed” might be a bit strong. Still, it is clear that AI is far from overhauling entire companies.

Yet, entry-level jobs have decreased.

I do not believe this is a contradiction. Even if firm-level innovation projects fail, AI as an auxiliary tool for individuals is already functioning well. Those auxiliary tasks are exactly what entry-level employees used to do: finding data, writing drafts, summarizing meetings, and writing simple code. Even if an innovation project stops, the work of one entry-level employee can still be saved.

Prescriptions to revive the entry-level hiring ladder: From cost support to retraining

Prescriptions emerging from seminars fall into two main categories.

For the youth, Professor Lee Chul-hee proposed cost subsidies.

His suggestion is for the government to cover a portion of corporate hiring and training costs to restore the skill-building ladder.

Regarding retraining, Yoo Jin-sung, a senior researcher at the Korea Economic Research Institute, identified skill enhancement and reskilling as key tasks. He proposed an employment service system that analyzes a worker’s job experience and capabilities to suggest transition paths.

Singapore’s “SkillsFuture,” a program that provides education credits to citizens over 25, was introduced as a model, along with calls to shift toward hiring and placement based on competency rather than academic credentials.

Meanwhile, at the August seminar, KDI researcher Han Yo-seop argued for increasing monetary compensation for unfair dismissal and enhancing internal flexibility, such as job reassignment, to lower labor costs while increasing the effectiveness of protection.

Evidence regarding productivity was also presented.

OECD senior economist Stein Brooker introduced the potential for AI to boost annual labor productivity growth by 0.4-0.9 percentage points over the next decade (Herald Business).

Personally, the part about cost support has been the most thought-provoking.

Companies not hiring entry-level staff means their cost calculations have already been finalized; the core issue is how much support would be needed to overturn those calculations. As far as reports go, neither the specific amount nor the method of support has been defined. The real issue in this debate moving forward will be whose money will be used to replace the bottom rung of the ladder.