Naver and Daum Accelerate the AI Search Agent Race

South Korea's two biggest players in web search, Naver and Daum, have announced new AI-powered services in recent weeks that aim to replace the classic list of results with links to a conversation with an AI assistant. Naver rolled out its conversational AI Tab to all users in late June, while Daum, owned by Upstage, launched a beta of its AI Summary service built on Upstage's own Solar model.

Naver built its new service on three technological pillars: a proprietary language model tailored to the company's products, an approach it calls harness engineering, and multimodal AI. The model is designed to handle long context and multi-turn conversations across the whole Naver ecosystem, including search, shopping, maps, blogs and user communities.

More small models instead of one big one

The key architectural shift is a move to a distributed system made up of multiple smaller models instead of a single large model handling every query. Naver says this approach doubles response speed and cuts operating costs by up to threefold compared with its earlier setup. The company also applied reinforcement learning, which it says cut hallucinations by up to 30 percent compared with the previous generation of its HyperCLOVA X model.

Over the coming quarter, Naver plans to add an AI Briefing feature, a visual search tool called Smart Lens, and real estate-related services to AI Tab. Later in the year, the company says it will bring AI agents to its own Whale browser and to health services, extending conversational AI beyond the search box into broader actions carried out on the user's behalf.

Daum bets on Solar and cited sources

Daum, historically South Korea's second-largest portal, came under the control of Upstage this year after the company acquired its operator, AXZ. The new AI Summary service automatically analyzes web documents and generates summaries with cited sources in categories such as finance, health, entertainment and trending topics. The service is currently in beta, with full rollout planned for later in 2026 alongside a conversational mode that will allow extended, multi-turn dialogue with the assistant.

AI Summary at Daum is a starting point that shows how AI models can bring real change - Kim Sung-hoon, CEO of Upstage

Part of a global search race

The moves by Naver and Daum fit into a broader shift away from the classic list of links toward conversational interfaces, a trend already underway at Google with its AI Overviews feature and at OpenAI, which has been building out search within ChatGPT. For the Korean market, where Naver has held a dominant position over Google for decades, the agent-focused approach to search is an attempt to defend market share against growing competition from global language models.

For Polish content creators and businesses dependent on search traffic, developments in Korea are a warning sign similar to the one already raised by Google's AI Overviews tests in Europe. If search engines shift en masse to AI-generated summaries instead of directing users to source pages, publishers and online stores should brace for further declines in organic traffic, even when those summaries include citations.

Naver has also said it will integrate AI agents with its Whale browser and health services later this year, meaning the real-world results of Korea's experiment with a multiple-small-models architecture and its push to cut computing costs should become clear within the next few months.

Sources: Naver, Daum ramp up AI search race with agent-focused strategies (koreatimes.co.kr)