Naver, Kakao gear up for agentic AI era in 2026

Naver's headquarters, top, and Kakao's office in Seongnam, Gyeonggi Province / Courtesy of Naver and Newsis

After spending last year overhauling their artificial intelligence (AI) business, Korea's two internet giants, Naver and Kakao, are now accelerating their push toward redefining their flagship search and messaging services through agentic AI.

Both companies spent 2025 strengthening the application of their AI technologies, embedding large language models (LLMs) across their ecosystems to improve user engagement in their existing services.

Building on these foundations, they are now positioning their AI systems to operate proactively as agentic AI, which can understand user intent, plan actions by itself and execute tasks without direct prompts from users.

In November, Naver unveiled Agent N, a personalized and integrated AI agent, as its core AI strategy to leverage the company’s vast data and translate it into real actions within Naver services. It is designed to seamlessly connect Naver’s services from search and maps to calendar, reservations and content.

“Users will no longer need to worry about which words to type for search. Instead, simply by conversing with Agent N, the AI agent will understand their intent, connect them to the desired content, products and services and even carry out real-world actions on their behalf,” Naver CEO Choi Soo-yeon said during the company’s Dan25 conference in November.

Screenshots of Naver's AI Briefing search results / Courtesy of Naver

Following a year of testing its on-service AI strategy, such as the AI Briefing feature, which now accounts for over 20 percent of search queries, the company plans to deepen AI's role across its core businesses.

Starting in the first quarter of this year, the company plans to roll out an AI shopping agent, followed by an AI tab within its search results in the second quarter. Agent N for Business will also be unveiled to support businesses and advertisers with automated recommendations and bidding optimization based on AI analytics.

To support agentic AI at scale, the company is also committing more than 1 trillion won ($691 million) to graphic processing unit (GPU) and AI infrastructure investments this year, while operating testbeds that connect its headquarters in Seongnam, Gyeonggi Province, and data center Gak in Sejong.

Screenshots of ChatGPT for Kakao / Courtesy of Kakao

Meanwhile, Kakao is focusing on building a closed-loop agentic AI ecosystem centered on its flagship KakaoTalk messenger, spending the past year on a company-wide reorganization to prioritize AI and improve profitability through KakaoTalk.

Early last year, Kakao formed a strategic partnership with OpenAI to bring global AI technology to its domestic user base of nearly 50 million. Building on the collaboration, it launched ChatGPT for Kakao in October, allowing users to access ChatGPT directly within KakaoTalk, while also accessing its affiliated mobility, payments, music streaming and gifting services through the interface.

Kakao is also refining its proprietary on-device AI model, Kanana, which proactively offers assistance based on conversational context. In the first half of 2026, the company plans to fully unveil agent-based Kanana in KakaoTalk and Kanana Search. These capabilities will later be integrated into major services, including Melon, Kakao Map, gifting and reservations, while an open-agent ecosystem with outside partners is also in the works.

“For AI to take the initiative in making suggestions, it is crucial to first understand the user’s situation and status, so Kakao has chosen on-device AI technology in order to provide diverse contextual information in the safest way and at the most necessary moments,” Kakao CEO Chung Shin-a said during the SK AI Summit in November.

“In particular, because Kakao has conversations, which carry the richest and clearest context, we have implemented systems that enable AI to independently infer what actions to take within that context.”

To support more advanced agentic scenarios, the company recently introduced Kanana-2, an upgraded model with enhanced tool-calling functions and improved instruction-following abilities.

The company will continue developing models optimized for complex agent workflows with a long-term plan to expand its AI into a multimodal model that can use not only text but also voice and images. It will also further lighten on-device versions, underscoring its ambition to build a fully integrated, conversation-to-execution AI ecosystem within KakaoTalk.