South Korea to Train Workforce to Verify AI-Generated Code in Agentic Era

South Korea’s Ministry of Science and ICT hosted a expert briefing on March 13 to discuss the “Agentic AI era, software industry and talent development,” chaired by Vice Minister Ryu Je-myeong. The event, held in Seoul, gathered policymakers, industry representatives and scholars to consider how AI that acts as an autonomous agent will reshape software development and worker training.

Experts said the focus for developers is shifting from writing code to verifying AI-generated code. They described “agency-enabled” or agentic AI as a step beyond generative AI, where AI systems act with a degree of autonomy and require human oversight, validation and debugging. While AI-assisted coding can boost productivity, it also demands better controls and verification practices.

Kaist computer science professor Seong Minhyuk underscored the changing role of programmers, arguing that the critical question is how to validate the code produced by AI models rather than how to generate it. In his view, the ability to assess and validate AI-derived solutions will become a core professional competence in the near term.

The discussion touched on broader U.S. industry developments as a point of reference. The session noted that Anthropic released its agentic AI product “Claude Co-Work” in January, with observers pointing to a steep revaluation of software firms’ market capitalizations in the United States. The implication for Korea is to study how U.S. industry shifts may affect global tech competitiveness and supply chains.

Some commentators challenged the notion that coding courses at top U.S. universities are disappearing. While courses teaching pure coding may be evolving, professors emphasized that foundational software engineering knowledge remains essential, and that teaching must address how to use AI tools responsibly and effectively to avoid issues such as AI misreporting or “hallucinations.”

Professor Seong stressed that fundamentals matter: the goal is not merely to teach tools but to cultivate judgment about when and how to apply them. He argued that AI should be used to augment humans, not replace core engineering thinking or oversight.

Song Ho-cheol, a senior executive at Doozeon Bizon, urged policy support beyond simply addressing automation fears. He called for government measures to advance AI-agent–based software development, including incentives and standards that encourage innovation while preserving human expertise.

In line with ongoing policy work, the Ministry has held six “SW industry and talent development innovation colloquiums” over the past month to refine its approach. The March meeting aimed to translate these discussions into concrete policy shifts, with participants calling for a workforce able to define problems, design systems, and verify results in AI-enabled software.

Officials and experts recommended prioritizing the cultivation of two talent streams: “advanced” software professionals with strong fundamentals, and “convergent” talents who combine domain knowledge with AI proficiency. A central emphasis was on building capability to debug and verify outputs produced by AI agents, not just to rely on automated generation.

Ryu Je-myeong closed by acknowledging the scale of change: “The AI tsunami is not just approaching; it’s already here,” he said, stressing that no single guide can steer everyone through the transformation. He urged broad, ongoing dialogue among industry, academia and government to chart a practical path forward for Korea’s software sector and its global integration.