Kakao is accelerating a strategy to “onboard 50 million users” to AI services as it pivots toward an agent AI era. Rather than relying on a single, massive model, the company plans to combine a lightweight top-level orchestrator with domain-specific agents to improve cost efficiency and scalability.
CEO Jeong Shin-a said on a conference call for Kakao’s 2026 first-quarter earnings that many agent AI services drawing attention today face limits, including excessive token consumption and privacy risks. “Those limits will be a clear opportunity for Kakao in the coming agent AI era,” she said.
Jeong described many web-based agent designs as “one huge agent” handling all domains and tools. She said screenshot-based vision-language models and computer-use agents can consume large numbers of tokens and may take longer to produce desired results.
By contrast, Kakao’s agent AI platform is designed so a lightweight orchestrator analyzes user intent while specialized agents collaborate. “We built it so tasks are carried out organically through protocols between agents,” Jeong said, calling it a structure that sharply reduces token use and processing time.
Kakao also previewed plans to disclose “Kanana 2.5,” a 150B-parameter model, following last year’s “Kanana 2.” Jeong said Kanana 2.5 was developed from scratch and optimized for the agent AI platform, showing stronger performance in execution-focused areas such as planning and function calling compared with leading global models.
She said Kakao’s in-house “Kanana tokenizer” also improved training costs and inference speed. Jeong said Korean can require up to three times more tokens than English when using general-purpose tokenizers. With the Kanana tokenizer, she said, Kakao confirmed up to a 40% reduction in training costs and up to a 60% improvement in inference speed.
This year, Kakao plans to expand user touchpoints, a key element of an agent AI ecosystem, with the goal of onboarding all 50 million KakaoTalk users through multiple service formats.
For users with higher AI familiarity and an intent to use paid services, Kakao pointed to “ChatGPT for Kakao.” For a more natural, everyday AI experience without charges, it highlighted “Kanana in KakaoTalk.”
Kanana in KakaoTalk uses conversation context to identify user needs and connect services from information to reservations and payments. Kakao said user activity has improved steadily after an iOS closed beta test and an Android launch, and retention has held at about 70%, similar to the closed beta.
Jeong said the service remains in an early stage, with work still needed to improve answer accuracy and quality and to refine when the AI should step in. She said Kakao will raise user satisfaction through model upgrades.
Kakao last month also introduced “Kanana Search,” a context-based discovery service. It provides place and product search results and trend-topic exploration inside KakaoTalk chat rooms. Jeong said that even in the third week after the beta launch, activity rose meaningfully compared with traditional keyword-driven search.
Jeong also cited momentum for ChatGPT for Kakao, which has surpassed 11 million cumulative sign-ups. She said monthly active users nearly doubled from the previous quarter, and monthly messages sent per user more than doubled, indicating the service is moving beyond one-time visits into repeated use.
“Kakao’s mid- to long-term AI vision is for all 50 million KakaoTalk users to have personalized agents,” Jeong said.
On advertising, Jeong said growing short-form and content consumption inside KakaoTalk is translating into stronger ad performance, making ads another key growth engine alongside AI. She said average daily valid short-form plays as of April more than doubled from shortly after the service launched, and demand for video-focused ads is flowing in quickly.
She said the revenue share from display ad products other than Bizboard expanded from the 10% range in the first quarter of last year to about 30% in the first quarter of this year, as the ad revenue mix diversifies beyond a single product and a small number of large advertisers.
Kakao said consolidated first-quarter revenue rose 11% from a year earlier, marking its first double-digit growth in eight quarters. Jeong said the company achieved “qualitative growth” by strengthening its core competitiveness.
Jeong said the second half will be an important turning point as users can broadly experience agents that start with domestic conversations and complete payments. She said Kakao will show a future in which all 50 million users communicate routinely with personalized AI agents on KakaoTalk.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.