The Daegu Metropolitan Government gave public notice on the 3rd of a proposed amendment to its rules on hiring public service workers and fixed-term workers. The authorized headcount for public service workers falls from 656 to 652, and three of the four positions cut are counselor posts. The counselor headcount goes from 42 to 39. Daegu said 39 people are actually working, so the adjustment reflects vacancies that were already there, and no one will be let go.
The adjustment draws attention because the same organization has been running AI consultation for some time. The 120 Dalgubeol Call Center has a system that converts what citizens say into text, analyzes what they are asking, finds an answer in a knowledge database, and reads it back as speech. This process of understanding sentences and picking answers is effectively the same work that human counselors did in the initial response. Even if it is a reflection of vacancies as Daegu explains, the reason the center runs without filling the empty posts requires an explanation.
The macro diagnosis comes out similarly from several directions. The Bank of Korea's report "AI and Changes in the Labor Market" found that 3.41 million workers in Korea, 12 percent of the total, are highly likely to be replaced by AI. It also projected that employment would fall and wage growth would be held down in jobs with high AI exposure indices. A report issued by the National Assembly Budget Office on the 26th of last month also shows a trend in which youth employment in accounting and bookkeeping, customer consultation, and writing occupations declined from November 2022, when ChatGPT appeared, through the first half of 2025, while computing, finance and insurance, and human resources management rose.
The Budget Office takes a step back here. While saying the evidence that the spread of AI directly caused the employment decline is not clear, it advised preparing for job volatility. It is a sentence that recommends preparation without settling causation. A government employment statistic showing that employment in professional, scientific and technical services fell this January is also read as a signal supporting that concern.
The counterarguments are substantial too. Lee Eun-taek, a researcher at KB Securities, noted that predictions of massive job losses came out during the Fourth Industrial Revolution as well, yet the unemployment rate in the 2010s actually went down. Yann LeCun, a professor at New York University, said at the AI summit in India last month that the discernment of today's AI is at the level of a cat.
Kim Se-hwan, a researcher at the same KB Securities, analyzed that stocks within the travel and airline industries are diverging over "fear of AI replacement." It is a phase in which the expectation of being replaced is priced in before replacement actually happens.
The market has been shaken hard by that expectation before. Citrini Research, a U.S. research firm, released a report on the 22nd of last month (local time) that assumed the year 2028. It warned that as ultra-high-performance AI tools spread, mass unemployment and mortgage defaults would break out at once, bringing "growth without consumption," in which the economy grows but spending does not follow.
On the day it was published, all three major New York stock indices fell, and the shares of the companies named were pushed down as well. IBM's share price fell the most since 2000. The White House dismissed the report as being at the level of science fiction.
Change on the tool side is fast as well. "Claude Cowork," the enterprise tool Anthropic released early this year, lets work such as document summarization, data analysis, and contract review be handled automatically even by people who are not specialists. After its release, a selloff broke out on the New York stock market centered on software companies, and the shock spread to cloud providers.
Shares of RELX, a data analytics company, fell as much as 31 percent on the New York Stock Exchange this year. The company's CEO, Dario Amodei, warned that half of white-collar jobs will disappear within five years, and in his January essay "The Adolescence of Technology" he wrote that the feedback in which AI fixes its own code and raises its performance is growing stronger.
The prescriptions split in opposite directions here. One side holds that the labor market must be made more flexible. In the country-by-country ranking of labor law flexibility issued by the Korea Industry Alliance Forum, Korea came last among nine countries with 43 points, falling short of the average of 70.61 points and also of Taiwan and China, each with 65 points.
The fact that GDP and GNI per hour worked are below the OECD average is another argument for this camp. The other side is arguing over whether the government should intervene directly in jobs, or move toward guaranteeing high income for everyone.
The data inside the labor market looks closer to movement than to disappearance. In JobKorea's analysis, AI-related job postings in 2024 rose 13.1 percent from 2022, and in a survey of 263 members, 79.5 percent said they use AI in their work. Over three years, applicants for AI positions jumped 1.7 times and competition per posting 1.8 times. By share of postings, AI/ML engineers came to 9.4 percent, web and app developers 9.2 percent, and data engineers 3.5 percent.
The compensation gap is clear as well. In the "HR Money Report 2026" JobKorea published on the 20th of last month, the highest-paying group among 21 major job categories was AI, development and data. The average is 49.47 million won.
Broken down by company size, it widens to 52.79 million won at large corporations, 44.83 million won at mid-sized firms, and 39.94 million won at small and medium-sized firms. On Wall Street, JPMorgan and others have mentioned large-scale staff redeployment, and since it is a method of changing the work people take on rather than pushing them out, it is in line with JobKorea's indicators.
Daegu's public notice will be finalized after gathering opinions. Three positions may not look like much, but as a case in which the procedure of recounting the human share at a public consultation window was left in a document, it becomes a precedent the next local government will refer to. Whether the voice that answers first when you call a call center is a machine or a person, and how many people are left to take over when the machine cannot find an answer, is the change citizens feel. The debate shows itself at windows like these before it is settled into policy.
