The Age of AI Agents
Begins in Korea

In March this year, KT launched its master AI brand, K Intelligence, taking the lead in building a uniquely Korean-Contextual AI ecosystem. What makes “Korean-Contextual AI” uniquely Korean and why is it important?

In essence, it means AI finely tuned to Korea’s language, culture, and societal context. KT defines the mission of Korean-Contextual AI as “an AI that best understands the needs of the nation, businesses, and individuals, and leads their AX (AI transformation).” To achieve this, KT established four core principles spanning infrastructure, data, models, and agents: Data Sovereignty (customer data rights must be strictly protected), Cultural Relevance (AI must deeply understand Korea), Full AI Model Line-up (customers should have the right to choose their preferred models), and Reliable & Responsible (AI must be safe, explainable, and compliant with regulations).
We’re currently preparing AI products built on these principles. Methodologically, KT also emphasizes open innovation. While leveraging our in-house technologies, we define global top players as collaborators rather than competitors, working together to deliver value to customers. Korean-Contextual AI is essential for enhancing national AI competitiveness, driving enterprise digital transformation, and ensuring the public can benefit from AI in everyday life.

KT is set to release a Korean-language large language model (LLM) co-developed with Microsoft later this year. Can you tell us more?

This model is based on GPT-4o but optimized for Korea’s linguistic and cultural landscape. It offers a seamless, contextualized AI experience by deeply reflecting Korean identity. It has already shown excellent performance in Korean-language understanding, generation, and instruction-following, as well as in handling social, cultural, and professional knowledge. We have not observed the performance degradation (catastrophic forgetting) often seen in fine-tuned models. Secure operation of this model will be implemented within KT’s K Infra (domestic cloud regional infrastructure) and offer customizable tuning environments for enterprises.
Our greatest competitive advantage lies in combining world-class AI capabilities with Korean-Contextual expertise, and KT plans to rapidly scale high-quality AI services across public and private sectors.

In 2024, KT took a surprising turn by announcing a partnership with Microsoft during a time when many companies were focused on developing their own AI. Given that KT launched its proprietary LLM, Mi:dm, in 2023, why pursue this collaboration?

Because of our clear vision of becoming an AICT company. While continuing to develop in-house capabilities through Mi:dm, we also want to rapidly bridge the technology gap with global leaders and create sustainable value by combining AI with cloud infrastructure.
The collaboration spans multiple areas—including the development of a Korean-Contextual AI model, secure public cloud (SPC), and AI-based digital transformation platforms. Mi:dm, our small language model (sLM), is being significantly upgraded for use across various industries. We are also leveraging external, global LLMs that are high-performing and technologically advanced to broaden the scope of Korean-Contextual AI. This transition to a multi-model strategy represents KT’s goal of increasing both quality and refinement in our AI service offerings. Microsoft’s expertise in sovereign AI and cloud infrastructure creates strong synergy with KT’s SPC and industry-specific AX strategies.

KT recently signed a strategic partnership with global big tech firm Palantir, set to take effect in 2025. What kind of synergies do you anticipate from this collaboration?

Palantir is a global leader in AI and data analytics, known for its unparalleled ability to support real-time decision-making in operational environments. Its decision support platforms—built on large-scale data integration and analysis—have proven so reliable and precise that even the U.S. government uses them in critical operations. By combining Palantir’s technological capabilities with KT’s telecommunications and SPC infrastructure, we can accelerate AX for government agencies and domestic enterprises in a secure and powerful way. Following our collaboration with Microsoft, this partnership further strengthens KT’s leadership by merging global tech excellence with our ability to execute on the ground.

KT recently formed an internal task force called “AXcelerator TF” led by employees in their 20s and 30s. How has this group driven a shift in your work culture?

AXcelerator TF is made up of younger employees who are directly involved in developing and using AI agents to revolutionize how we work. One of their first initiatives was to centralize all company documents on the cloud, enabling AI learning and agent-powered search. This required a full overhaul of our internal infrastructure and was a cultural challenge, but the payoff has been significant. We’ve moved beyond sharing documents via email and now collaborate in real time, boosting productivity. AI agents have been particularly effective in document analysis, summarization, and data extraction.

What other internal changes have you seen from adopting AI agents?

KT provides Microsoft Copilot training and tools to all employees—regardless of their technical background—so they can build and use AI agents relevant to their roles.
This has improved both AI literacy and application skills, fostering a culture where AI is naturally seen as a “digital collaborator.” Now, departments like HR, finance, and legal are actively working with AI agents, leading to improved productivity and a growing AI-enabled culture within the company.

What are some key elements to consider when introducing and applying AI agents to their operations?

It’s a misconception to define AI agents as just an automation tool. Also, driving real innovation requires fundamental redesign of processes. This includes data governance, IT integration, UX, security, performance, and change management. Key obstacles—such as data privacy, system integration, and implementation costs—require meticulous planning. You should also address often-overlooked issues like AI judgment errors, employee resistance, and algorithmic bias by establishing ethical principles and response systems.
Ultimately, success comes from a holistic redesign at the organizational level—integrating people, technology, and workflows in unison.

Could you briefly introduce Wrtn Technologies and its services?

Since our university days, my co-founders and I have been driven by a mission to help people push beyond human limitations. In April 2021, we founded Wrtn Technologies with the belief that generative AI could bring this vision to life. Today, Wrtn is one of the fastest-growing consumer AI platforms in Asia. Beyond core productivity tools like search and writing, we offer a diverse range of experiences, including AI-powered character chats designed for everyday enjoyment.

One in ten Koreans now uses Wrtn AI—an impressive feat in under two years. What fueled such rapid growth and innovation?

From the outset, we believed that generative AI would become a utility as essential and ubiquitous as the internet, electricity, or even air. This belief led us to embrace a guiding principle: “One person, one AI.” Achieving that goal meant focusing on two key factors: accessibility and practicality.
For accessibility, we made an early and bold decision to offer our core services entirely free of charge. Today, Wrtn is used across government agencies, schools, and community centers nationwide.
For practicality, we knew AI had to go beyond productivity to truly enhance users’ lives. In April, we launched Wrtn 3.0, which introduced features like AI Assistants and App-Tech—earning opportunities built into the app—that have been widely embraced by users.

With AI platforms rapidly evolving—Meta’s open-sourcing of LLaMA and Anysphere’s Vibe coding tools, for instance—what is Wrtn’s positioning strategy?

At Wrtn, we’re committed to enabling all businesses to transition their existing services into AI-powered agents. That’s why we’ve open-sourced key projects, including our new framework Agentica, launched in April. Agentica addresses one of the biggest pain points in agent development: high error rates when integrating the model with external systems. Such high error rates occur even in top-tier LLMs due to incorrect selection of tools or parameter handling errors. Thanks to Agentica, developers can now build stable and effective AI agents that operate independently—without constant human correction. As a leading AI company in Korea, our goal is to foster a hassle-free ecosystem where anyone can develop AI agent systems.

What role can cooperation between large corporations and startups play in the AI ecosystem?

I remember when we first gained early access to GPT-3 to build our initial generative AI service. Many doubted we could succeed, citing high costs, hallucination risks, and quality issues. However, we were confident that these issues would be resolved naturally with AI model advancements. Instead, we focused on upgrading our services. By the time ChatGPT launched in November 2022, many such issues indeed resolved themselves as the technology matured. Likewise, when we made our services completely free, some questioned our revenue model. However, we stayed the course believing that the costs of utilizing AI would drop dramatically. That persistence paid off—today, Wrtn has surpassed 5 million monthly active users.
Startups are constantly fighting an uphill battle against skepticism, but this is not a lonely fight. Companies like Naver and Kakao paved the way, choosing to be first movers despite uncertainty around cost, monetization, and global competition. I believe startups that take bold, early actions when entry barriers are low may lead major transformations.