Korea's 'Everyone's AI' Won't Just Chat — It Will Make Calls and Pay Bills

Korea Is Building an AI That Calls Hospitals Instead of Answering Questions

Most countries shipping AI to consumers right now are perfecting the chatbot — a system you type into, that replies, maybe sends you a link. South Korea is trying something different. It is asking AI to pick up the phone.

On September 4, the Ministry of Science and ICT unveiled three consortium prototypes under its “Everyone’s AI” (모두의 AI) initiative, and the most striking demo came from SK Telecom: an AI agent that searched for evening-clinic availability, then actually called a hospital, listened to the automated response, learned there were two patients waiting, and — after the user confirmed — sent a ride route via T-map. No app installed. No form filled. A text message initiated the whole chain.

That is the difference between an AI that talks and an AI that acts. The Korean government is explicitly choosing the latter, and it is backing the bet with 512 GPUs this year and a 250 billion won allocation next year alone. Beta launches in September. Public rollout follows in December.

The Three Consortia Are Building Different Doors Into the Same House

The project brings together the country’s three largest telecom and platform ecosystems. Their strategies converge on one goal — execution, not explanation — but their entry points diverge in ways that reveal distinct assumptions about who gets to use AI first.

Kakao is treating KakaoTalk — a messaging app with nearly 49 million monthly active users in a country of 52 million — as the operating system for personal AI. Its “1-person N-agent” concept envisions a chat interface where multiple specialist agents handle everything from document searches to payment in a single conversation thread. LG U+ provides the voice layer for phone-based access. Kakao projects 5 million MAU at launch, climbing to 10 million by end of 2027.

KT is building a service aggregation platform called “I-eom” (이음) that stitches together 12 daily-life categories — commuting, shopping, pensions, education, finance — through a domestic model it calls “token factory,” which dynamically routes each query to the cheapest capable model. The pitch is cost efficiency plus coverage: a first-job entrant gets outfit recommendations and purchases; a small business owner gets vehicle inspection reminders and booking; an elderly person gets notified before their pension eligibility date with the application pre-filled.

SK Telecom is going the furthest on the action side. Beyond the hospital-calling demo, it plans to build an employment-coaching agent that finds job postings, surfaces similar opportunities at competing firms, drafts cover letters and conducts simulated phone interviews. Crucially, SKT is designing its interface around people who do not own smartphones: a phone call or text message triggers the entire workflow. For a country where the elderly population exceeded 18 percent and smartphone literacy among those over 70 remains modest, this is not an afterthought — it is the core design constraint.

The Real Risk Isn’t Technology. It’s Indifference.

Minister Bae Gyeong-hoon did not mince words at the September briefing. Korea already uses foreign AI services extensively. If Everyone’s AI cannot produce a clear differentiation from ChatGPT, Claude or Gemini, “it will fail,” he said. He demanded a “wow” moment at December’s launch, not incremental improvement on existing models.

That framing matters because it exposes the project’s central tension. Korea has world-class telecom infrastructure and deep platform penetration. It has also spent years watching foreign AI companies capture user attention and data with products that work well enough for most tasks. The question is whether a domestically built execution layer — even a well-funded one — can overcome the network effects already stacking behind American and Chinese platforms.

The government’s answer appears to be: localization plus access. Every consortium emphasized that their models are trained on Korean language, cultural norms and administrative workflows. A foreign model might understand the word for “resident registration certificate,” but it probably cannot navigate the actual submission process through Korea’s complex municipal portals. That gap — the space between understanding a language and executing within a bureaucratic system — is where Everyone’s AI hopes to build a moat.

What This Means for Southeast Asia

The implications extend well beyond the Korean Peninsula. Several ASEAN countries share Korea’s demographic profile — rapid aging, high mobile penetration, underdeveloped digital-government infrastructures — and none has yet mounted a coordinated national push of this scale.

Vietnam, Thailand and the Philippines are all courting foreign AI providers with favorable regulations and tax incentives. Korea, in contrast, is building its own execution stack and making it accessible through the channels its citizens already use daily. That approach could become a template: rather than importing capability, a government partners with dominant telecom-platform combinations to ship AI through existing user relationships.

The 500-million-user target cited in project discussions likely refers to a broader ambition across exportable markets, not just domestic adoption. If Korea can prove that a state-coordinated AI rollout reaches elderly, low-literacy and non-smartphone users at scale, it creates a product narrative that several ASEAN governments might find attractive — particularly those wary of ceding data and administrative access to foreign cloud providers.

Who Wins and Who Loses

The clear winners in this configuration are the three consortium anchors: SK Telecom, Kakao and KT each gain a deepened relationship with their user base and a data flywheel that grows every time an AI completes a real transaction. Their competitors in messaging and telecom — notably Naver, which was notably absent from this announcement — face a defensive squeeze.

Foreign AI providers face the harder calculation. OpenAI, Google and Anthropic already serve Korean businesses. Competing for consumer-facing execution in a market where the government is subsidizing domestic alternatives with GPUs and direct procurement will require either deeper localization or a willingness to accept a narrowed role.

The everyday user is the swing variable. If the December launch delivers genuinely useful automations — not clever demos that break on the first edge case — Korea may prove that a national AI strategy can reach further down the socioeconomic ladder than the current generation of consumer AI products. If the prototypes fizzle in production, the 250 billion won commitment will look like expensive theater.

Minister Bae’s demand for a “wow” moment is both a challenge and an admissions test. Korea has the infrastructure. It has the platforms. What remains to be seen is whether execution beats conversation when users get to choose.