SEOUL, May 7 (AJP) - Near midnight in Seoul's Gangnam district, the lights inside law offices still burn long after neighboring buildings have gone dark. But for many junior lawyers, the exhaustion no longer comes from endless paperwork alone.
Artificial intelligence has sharply shortened the time needed for much of the grunt work that once consumed young associates' nights — document reviews, precedent searches and first-pass drafting. Yet the workload itself has not eased. Many firms, increasingly reliant on AI tools, are simply hiring fewer junior workers to handle what remains.
For 37-year-old law firm assistant James Kim, complaining is hardly an option in a profession where even landing an entry-level position has become fiercely competitive.
"The fear now isn't overwork," he said. "It's eventually not being needed."
Across South Korea's white-collar industries, AI is beginning to reshape not only how young professionals work, but whether they are hired at all.
As firms automate more of the routine tasks once assigned to junior employees, younger workers increasingly worry that AI is eroding the apprenticeship system through which future experts were traditionally trained.
The anxiety is becoming increasingly visible in employment data.
According to the Bank of Korea, nearly 98 percent of the roughly 211,000 youth jobs lost between July 2022 and July 2025 were concentrated in industries with high AI exposure. Employment in South Korea's professional, scientific and technical services sector fell by about 105,000 in February from a year earlier, marking the steepest decline since the current industrial classification system was introduced in 2013.
For many younger professionals, the labor market increasingly feels less like the beginning of a career than a struggle for survival.
"Law school graduates attend bar association programs during the day and spend hours at night sending out resumes," Kim sighed.
The shift is exposing a growing paradox of the AI era. The same tools boosting productivity may also be weakening the apprenticeship structure through which junior workers traditionally became experts.
"These tasks were not glamorous, but they were often the training ground through which young professionals learned how the business works," said Kwon Hyeok-koo, a professor of Nanyang Technological University's business school.
For decades, junior employees learned through repetition — researching cases, drafting reports, making mistakes and receiving corrections from senior colleagues. AI is now compressing or bypassing many of those stages altogether.
According to Yoo Joshua, a lawyer at a Seoul-based law firm, younger lawyers are losing the process through which they once learned "how to think" in practice as AI has taken up much of their work.
The longer-term risk is not simply job displacement but erosion of professional judgment itself.
Lyse Langlois, a professor of Industrial Relations at Laval University and director of the International Observatory on the Societal Impacts of AI and Digital Technology (OBVIA) said younger workers risk developing what she described as an "autopilot" relationship with AI — relying on AI outputs passively without critically questioning or verifying them.
"If people stop critically examining AI-generated outputs, they may gradually stop developing independent judgment altogether," she warned.
Kwon said reducing junior hiring may appear efficient in the short run, but could damage the future supply of skilled professionals. The key challenge, he said, is whether younger workers can still develop the judgment needed to recognize and correct AI's mistakes.
Citing the "jagged frontier" findings from a recent field experiment by global consulting firm Boston Consulting Group, Kwon said AI can improve performance within its capability range but may also increase errors when users overtrust plausible but incorrect outputs.
As a result, he said, early-career skills are shifting toward AI orchestration, verification and judgment under uncertainty.
"It changes the role of junior workers from 'doing the first version' to 'supervising, checking, and improving the machine-generated first version,'" Kwon said, noting that entry-level professionals now need different skills, including knowing what questions to ask, recognizing flawed outputs, understanding company-specific data and judging whether AI-generated answers make sense in context.
Kwon also warned, "AI can widen gaps when the valuable part of work depends on tacit knowledge, client relationships, reputation, or contextual judgment."
The transformation is already reshaping hiring patterns across industries.
Kwon, a 37-year-old architect who now leads an AI-related division at his firm, said generative AI tools have radically accelerated conceptual design work over the past several years.
"Tasks like generating images or exploring design alternatives have become dramatically faster," he said. "Tools like Midjourney and Nano Banana — AI-based image and concept visualization tools — allow architects to visualize multiple design concepts almost instantly."
Yet he said the final stages of professional work remain deeply human.
"You still need people to refine prompts, interpret client demands, apply structural logic and ensure compliance with regulations," he said. "AI can generate options, but judgment remains a human responsibility."
"In architecture, entry-level hiring has dropped by nearly 50 percent, while demand for workers with five to 10 years of experience continues to grow," he said.
"Today, firms care less about pure design ability and more about whether someone can strategically use AI tools while coordinating across multiple disciplines."
Kwon of Nanyang argued that firms using AI mainly as a cost-cutting tool may boost short-term efficiency while weakening long-term talent development.
"The bigger risk is that it erodes the apprenticeship model through which young professionals become experts," he said. "Firms that redesign training around AI will benefit; firms that only cut junior tasks may face a future talent shortage."
At the education level, Kwon stressed the need for universities to adapt their training models as AI becomes embedded in professional work.
"Universities need to teach AI literacy together with domain reasoning," he said. "Students should learn not only how to use AI tools, but also how to challenge them, audit them and combine them with human judgment."
He also warned that Korea faces additional structural risks because adult-learning participation remains relatively low. Citing OECD data, Kwon noted that Korea's adult-learning participation rate stands at about 13 percent, compared with an OECD average of 40 percent, underscoring the need for stronger lifelong-learning systems, reskilling incentives and support for smaller firms with limited training resources.
Erik Cambria, professor of Artificial Intelligence at Nanyang Technological University (NTU) similarly argued that future education systems will need to prioritize conceptual understanding over task execution.
"The value in the emerging 'relationship economy' increasingly comes from understanding context, intent and human nuance rather than simply producing outputs," he said. He added that education and workplace training should place greater emphasis on critical thinking, interdisciplinary understanding and the ability to work with — and critically assess — AI systems.
Not every profession, however, is experiencing AI as a direct threat.
In medicine, AI is increasingly seen as a tool that reduces repetitive work while allowing doctors to focus more on diagnosis and their core responsibilities.
South Korea's medical and social welfare sector added 294,000 jobs in March from a year earlier, the largest increase among all industries, sharply contrasting with declines in technical fields where employment fell by 61,000 during the same period.
He said the impact varies by specialty.
"Radiology and emergency medicine are seeing the fastest changes because AI can quickly assist with analysis," he said.
"Tasks like organizing patient records, chart documentation and scheduling are becoming increasingly automated, which allows doctors to focus more on judgment and treatment."
Data-heavy departments such as pathology and cardiology have also seen major efficiency gains, he added. By contrast, surgical specialties and psychiatry — where face-to-face patient interaction remains central — have experienced relatively limited disruption.
The shift is increasingly reflected in broader industry research as well.
According to the Anthropic Economic Index, AI use is increasingly centered on "augmentation" — working alongside humans — rather than full automation replacing them.
The report found that while AI is reducing repetitive tasks, it is simultaneously increasing the importance of higher-level responsibilities such as management, contextual judgment and decision-making. In effect, researchers said, AI is driving both "deskilling" in routine work and "upskilling" in more complex, judgment-intensive roles at the same time.
Experts say adapting to that transition will require a redesign of education and workplace training around an "AI apprenticeship model."
Rather than banning AI-assisted work, Kwon said educators should require students to explain what prompts they used, what outputs they verified and why they accepted or rejected certain results.
He said organizations still need to preserve some foundational work without AI so junior employees can understand the underlying logic of their profession.
"At the same time, managers should create AI-assisted assignments where juniors must explain what the AI produced, what they accepted or rejected, and why," he said, adding that "managers should evaluate the process, not only the final output."
He also said companies should require junior workers to document how they used AI, including what prompts they entered, what assumptions they checked, which sources they verified and what risks remained. According to Kwon, this would turn AI from an invisible shortcut into part of the learning process itself.
Kwon further emphasized the need for more structured mentoring systems, warning that while AI can make younger employees more independent, it may also reduce interactions with senior colleagues and make it harder for managers to detect knowledge gaps.
"Firms should rely less on informal help-seeking and instead introduce more deliberate check-ins, case reviews and post-task debriefings," he said.
Langlois raised what she called a more fundamental question.
"What do we want to value as a society?" she asked. "A model driven by hyper-competitiveness and innovation, at the risk of leaving a significant portion of the population behind, or one in which everyone has a place, and where AI serves human beings?"
Shortly before 2 a.m. in Seoul's Gangnam district, the lights inside an architecture office remained on.
AI-generated renderings filled one side of a monitor. Structural calculations covered the other. Coffee cups sat stacked beside keyboards.
The output was arriving faster than ever.
But judgment — and responsibility for the outcome — still rested with humans.
Copyright ⓒ Aju Press All rights reserved.