GAIF 2026: Kakao, SK Telecom and Naver Cloud unveil actionable AI plans

Kakao, SK Telecom and Naver Cloud presented their actionable AI plans on Thursday (12) in Seoul: systems that do not just answer questions, but actually carry out tasks such as placing orders, making bookings and processing payments. The three companies detailed their AI transformation (AX) strategies at the 13th Edaily Global AI Forum (GAIF 2026), held at the JW Marriott Dongdaemun.

The shared message was clear: competition in AI is shifting from a contest over model performance to a contest over real execution capability.

Kakao: an agent that remembers — and asks permission to act

Kakao put a hyper-personalized agent at the center of its presentation: one that remembers users' tastes and context and directly handles ordering, booking and payment. The company's chief technology officer, Song Jae-ha, drew a distinction between fact and interpretation: "A purchase is a fact; a preference is an interpretation." Judging preferences from purchase records alone, he warned, can lead to repeated off-target recommendations — such as learning that children's shoes bought as a gift for a niece or nephew reflect the buyer's own taste.

To address this, Kakao is pursuing Project Mosaic, which separates and remembers behavior and preference: confirmed facts such as purchases and searches are stored separately from tastes the AI has interpreted, and users can correct the basis for judgments. For the same request — "find me a restaurant for tomorrow evening" — the recommendation changes with context: college friends, family, or a vegan friend. "An AI that remembers me must also remember that I can change," Song said.

The service touchpoint is KakaoTalk. The company is preparing Kanana Space, which lets users summon the AI with a two-finger swipe up from any screen and assign tasks by voice or text. Combined with an "AI conductor" that links services such as search, maps and reservations in sequence, the system is designed to handle shopping, booking and payment in one go. On the in-house model front, Kakao has already developed Kanana-O, which understands images, voice and text, and is developing Kanana Live, which listens and speaks at the same time.

Kakao's guiding principle is the separation of memory from execution authority. "Memory is the basis for judgment, but it is not the authority to act," Song said. "Knowing what I bought last time doesn't mean it's okay to order at any price." For hard-to-reverse actions such as payments, the system is designed to require the user's explicit approval. "Now failures will come through actions, not words," he said, explaining that the company is strengthening safety systems that test execution plans and external tool calls in a test environment, since a wrong answer can lead to a real ordering error.

In parallel, Kakao is expanding its ecosystem: "AI for Everyone," which Kakao leads, involves 34 companies and institutions, and about 500 tools are registered with "Play MCP," which connects external functions. The company is also preparing "Canvas," which lets individuals form agent teams on a PC and assign tasks via KakaoTalk. The ultimate goal is "one agent per person," with each user directly setting the scope of memory and execution authority.

SK Telecom: AI by phone call and text message

SK Telecom focused on lowering barriers to entry. Kim Ji-hoon, head of the company's AI business division and an executive vice president, said the criteria for AI competition are shifting beyond model performance to economic viability, technological sovereignty and the ability to perform real work. The company is preparing a way to use AI agents through phone calls and text messages alone, without installing a separate app or logging in. The idea is to let even people unfamiliar with digital services use AI with just a phone number.

The carrier also presented cases that go beyond one-off information searches: one example is a "job coach" that finds job postings suited to the user and helps with everything from writing applications to mock interviews.

Naver Cloud: sovereign AI for defense

Naver Cloud presented a strategy to apply sovereign AI to the defense sector, which is directly tied to national security. Yoo Kyung-bum, an executive director at Naver Cloud, stressed that in military intelligence and operational systems, securing control over data and infrastructure — not just AI models — is essential. AI can be used for battlefield intelligence analysis, support for commanders' decision-making, drone and satellite imagery analysis and the operation of surveillance systems, but relying on external technology carries risks of information leaks or service disruptions.

That requires specialized models trained on internal military data and secure computing infrastructure. Command headquarters need high-performance models for large-scale data analysis, while the actual battlefield requires lightweight models that run quickly. Yoo estimated that providing AI services to service members could require about 1,000 graphics processing units (GPUs), and advanced closed-circuit (CC) TV image analysis could require about 3,000. "We need to combine private-sector AI technology with military data to create a demonstration environment, and build an ecosystem in which defense companies and startups participate together," he said.

From performance to execution — and new risks

The three presentations show the same shift: the competitive axis of the AI industry is moving from model performance to real execution capability. Kakao is seeking differentiation in everyday ordering and booking, SK Telecom in popularization through telecom touchpoints, and Naver Cloud in the public and defense sectors where security is critical. Industry observers expect that as agents move to the stage of actually handling payments and work, the damage from malfunctions, the question of liability and the securing of data control will become core competitive strengths.

Context on Geekiki

Geekiki has already followed Kakao's earlier steps in this direction: on November 5, the company confirmed the Kanana Space pilot for December, aiming for 10 million monthly AI users by the end of 2026 (read here), and on November 6 it introduced the Kanana-2-v-decide decision model (read here). GAIF 2026 is the next chapter of that strategy — now with SK Telecom and Naver Cloud showing their own routes to the same idea: AI that executes.

Image credit: Kakao logo (the company's corporate identity), via Wikimedia Commons.

Sources

Transparency: this article was produced with the assistance of artificial intelligence, cross-checking public posts and internet sources. Please always confirm the information with the original sources.

Muse