Korea charts agentic AI era; emphasizes human design and upskilling.
A government-hosted roundtable on the era of agentic artificial intelligence and Korea’s software industry and talent development was held on the 13th in the main conference hall of the Kyobo Building in Gwanghwamun, Seoul. The event was organized by Korea’s Ministry of Science and ICT and chaired by the National Science and Technology Advisory Council, with industry and academic participants weighing the implications of advanced AI for software work.
The forum highlighted a widely discussed question: even as AI systems can write code, human roles in software remain essential for design and verification. Song Ho-cheol, head of the Medical Intelligence division at Duzon Bizon, a major Korean ERP software provider, said AI is reshaping how software work is done as capabilities improve, and that developers’ tasks are evolving accordingly.
Duzon Bizon was cited to illustrate a broader trend. The company specializes in enterprise resource planning software, and the discussion cited recent moves in the global AI space, including Anthropic’s January release of Claude Co-Work, which some observers linked to a decline in U.S. software stocks amid fears of a “SaaS Apocalypse” as AI enables new automation.
A key concept discussed is “vibe coding” or agentic AI coding, where natural-language prompts guide AI to generate code. Proponents say this approach can boost development speed, improve collaboration between non-technical and technical staff, lower entry barriers, automate repetitive tasks, and allow developers to focus on higher-value work such as validation and system optimization. But speakers stressed that AI is not a turnkey solution and requires careful oversight.
The roundtable also acknowledged clear limits. AI tends to struggle with complex software architectures and long-term maintenance, can produce hallucinations, and raises security and data-leak concerns. Song cautioned that while AI excels at implementing individual functions, it has difficulty with architectural design and structural judgments that support scalable, consistent systems over time.
A broader takeaway is that the distribution of coding work could increasingly hinge on domain knowledge. The idea is that people who understand tax law, human resources, and other specialized areas will increasingly guide how AI writes and tests software, effectively steering the collaboration among multiple AI agents rather than relying on coding prowess alone.
Education and talent development were another focus. Industry participants said AI adoption is softening new-hire demand in some sectors, underscoring the need to shift university curricula away from pure coding toward design, verification, and domain understanding. Stanford University in the United States runs a course on “modern software development” that emphasizes how to work with AI rather than how to handcraft every line of code. KAIST professor Sung Min-hyuk, who attended the event, cited the idea that students still must solve the “last 20%” of problems—where humans are needed—even if AI writes about 80% of the code.
Education equity emerged as a concern tied to capital. Prof. Sung noted that some students may achieve better results simply by spending more on AI tools, a gap that could widen with unequal access to resources. The discussion pointed to the need for policy and institutional responses to ensure that AI-enabled productivity gains in Korea—whether in tech, manufacturing, or services—do not exacerbate disparities or leave some talent behind. For U.S. readers, the exchange signals how North American tech firms and universities might prepare for a similar transition: balancing AI-driven efficiency with rigorous design and verification, while safeguarding security and training the next generation of workers beyond traditional coding.