South Korea weighs agentic AI's impact on software, reshaping talent policy

Seoul is weighing how to adapt as agentic artificial intelligence transforms software development. In a March 13 meeting, South Korea’s Ministry of Science and ICT discussed the rapid evolution of AI that can plan, analyze data, and write code, using the term “SaaSpocalypse” to describe the potential end of traditional software-as-a-service models.

The event was held in the main conference hall of the National Council for Science and Technology Advisory in central Seoul’s Gwanghwamun area. It followed six prior “Colloquia on Software Industry and Talent Innovation” over the past month, gathering roughly 70 experts from industry, academia, and government to refine policy directions.

Experts warned that agentic AI could dramatically speed up software projects, with some analysts estimating that tasks once taking three years could shrink to about 40 days. Such acceleration would force a fundamental rethinking of how software is built and who builds it.

Cho Jun-hee, president of the Korea AI and SW Industry Association (KOSA), said customers are shifting from buying packaged software to using AI to develop only the features they need. That shift, he argued, signals a move from suppliers to buyers and calls for a comprehensive reorientation of talent policies.

Industry observers noted changes in how development teams are organized, with the traditional one-project-for-one-development-team model giving way to new structures as AI handles more of the coding work. The human role is increasingly centered on defining requirements and guiding AI-driven processes.

At the forum, Song Ho-cheol of Duzon Bizon stressed that as AI handles coding, developers spend more time designing how AI should operate, signaling a fundamental change in the developer’s job. Shin Jeong-kyu of RabbleUp added that the field is shifting toward defining and validating the system, and urged universities to prioritize fundamentals like system programming and architecture over basic coding.

Wang Ji-young, a project manager at Kakao Enterprise, suggested strengthening cross-department collaboration and project-based learning rather than creating new departments, so students gain hands-on experience in solving real-world problems. There were also calls for practical infrastructure to support AI education, including GPU clusters.

KAIST professor Seong Min-hyuk highlighted a key bottleneck: Korea lacks school-level GPU clusters for AI training and education. He proposed creative solutions such as repurposing older enterprise GPUs and building shared, multi-university clusters to lower costs and expand access.

Ryu Je-myeong, the second vice minister of MSIT, announced that the government intends to finalize a comprehensive AI-era talent strategy in consultation with experts and to overhaul the AI workforce support system to better match on-the-ground demand and an AI-native environment.

The Korean discussions come as U.S. policymakers and industry observers watch the global shift toward “agentic” AI. In January, Anthropic released Claude Co-Work, which some analysts say could accelerate enterprise AI use and influence software markets, contributing to a noted swing in the market value of major American software firms by about $1 trillion. For U.S. readers, the Korean debate highlights a shared priority: ensuring a skilled, adaptable workforce to compete in a rapidly changing tech landscape, while considering how education, infrastructure, and policy must evolve to keep pace with AI-enabled productivity gains and supply-chain resilience.