if(kakao)26: Shifting Focus from Conversational AI to Agentic AI

Details

At the if(kakao)26 conference, four key technical shifts were discussed as core themes: moving from conversational AI to goal-completion agentic AI, native multimodality, purpose-specific infrastructure orchestration, and ensuring system-level reliability.
The event showcased agent-to-agent (A2A) collaboration cases such as Kanana Space and Workmate, and unveiled a multimodal model lineup including Kanana-o and Kanana Live. It also emphasized efficient resource management through SLMs and on-premises GPU-based LLMOps.
KakaoPay presented the importance of managing execution permissions and establishing quality gates through batch control, AI output verification, and LLMOps sessions. In the presented case study, the number of defects detected during the verification process increased from 5 to 8.
At the organizational level, it was proposed to convert individual AI usage into shared organizational assets and to build a quantitative performance evaluation system for repetitive task automation through internal AI platforms and system integration.
Technical sessions introduced cases of achieving a 96% cost reduction through the automation of over 150 experiments and the application of ultra-lightweight models, as well as the application of agent harnesses to the migration of over 200 legacy JSPs.

Key point