South Korea's Ministry of Science and ICT held a conference on May 13 titled "Agentic AI Era: Response Strategies for SW Industry and Talent Development," bringing together industry and academic experts.
Participants agreed that as artificial intelligence now handles most software development tasks previously performed by humans, talent cultivation systems must also change accordingly.
The conference was organized to develop concrete government policy shifts for AI and software based on discussions from six "SW Industry and Talent Development Innovation Colloquium" sessions held over the past month.
Agentic AI refers to autonomous artificial intelligence that independently formulates plans, analyzes data, and utilizes tools to complete tasks toward set goals without detailed human instructions. Unlike generative AI that merely provides information or answers, its core function is acting as a human "agent" by making judgments and taking actions autonomously.
Experts noted that productivity has improved dramatically with agentic AI, with software development projects that once took three years now completed in 40 days. Some analysts suggested the industry is entering an "SW production acceleration phase" where costs approach zero as production shifts from humans to AI.
Anthropic, developer of Claude, defined this year as "the year SW development evolves into interactions between AI agents," forecasting fundamental changes in the software industry structure. This has heightened the need to redefine AI and software talent development paradigms.
Song Ho-chul, Division Head at Douzone Bizon, who delivered a presentation, stated, "Advanced large language models' automated code generation now enables software implementation through natural language instructions alone." He added, "As so-called 'vibe coding' spreads, the scarcity value of traditional coding skills is declining."
Consequently, developers' roles are shifting toward maintenance and verification of complex AI-generated code. Song noted, "Organizations will see planning and judgment tasks outweigh execution-focused work. When hiring, broad domain experience, AI tool proficiency, and abilities to provide directional guidance and review core logic in SW implementation will become increasingly important."
Sung Min-hyuk, Professor at KAIST's School of Computing, presented on "Computer Science Education in the Vibe Coding Era." Drawing on discussions with Stanford Professor Mehran Sahami, who gained attention for his "Modern Software Developer" course, Sung offered perspectives on education in the AI era.
"Students who learn to read, debug, and structure AI-generated code rather than relying on simple prompts are more successful at programming," Sung said, emphasizing the importance of such capabilities.
He added, "Future computer science curricula will prioritize foundational competencies such as problem definition and structuring, system design, and debugging and evaluation over code writing itself." Given the rapid evolution of AI models, he advised focusing on developing "judgment" rather than teaching specific tool usage.
Conference discussions also proposed that the government should support universities' AI and software education redesign by providing core infrastructure including industrial and public data, GPUs, and servers, while strengthening project-based practical training.
Vice Minister Ryu Je-myung of the Ministry of Science and ICT stated, "In the agentic AI era, we are working with experts to newly define the talent profile needed in the field and comprehensively reviewing existing AI workforce development and support systems." He added, "We will identify core AI competencies needed by advanced professionals, practitioners, and job seekers alike, and establish comprehensive talent development policies befitting the agentic AI era that can respond to field demands and AI-native environments."