*This content was translated by AI.
On the 17th, KT held a press briefing at the KTGwanghwamun Building in Jongno-gu, Seoul, revealing that it is researching 'autonomous agents' and unveiled its blueprint for the agent AI business.
Autonomous agents are AI services that directly perform tasks such as clicking buttons and moving files on users' PCs or in the cloud, with Openclaw being a representative example. Existing autonomous agents raised concerns about malfunctions, such as the leakage of personal information like passwords and unauthorized deletion of emails, due to granting AI excessive control authority. There was also a limitation that they were difficult to handle without relevant knowledge.
KT is developing a Korean version of Openclaw to address these issues. The ultimate goal is to launch Employee agents that collaborate and control each other.
Kim Jun-seok, Head of KT Agentive AI Lab (Senior Executive), stated, "Employee agents will begin pilot use within the AX Future Technology Institute next month before expanding company-wide. They can also be sold externally to the government, military, and other companies." On this day, KT released a video showing four autonomous agents scheduling a golf appointment and an agent responding on behalf of a user on vacation.
Additionally, KT plans to commercialize the B2C service 'Hyper-Personalized AI Agent' in the second half of the year. This involves enhancing existing services such as My K, Genie TV, and Sajangiji by integrating long-term memory and actionable agent technologies. For example, the My K plan management app will analyze user patterns to recommend the optimal plan and proactively notify users of unused benefits.
In the B2B (business-to-business) sector, KT will develop 'Vertical AI Agents' optimized for specific industries. These include network agents that create response manuals when network outages occur and patent agents that determine patent application eligibility and draft declaration forms. KT aims to prove practical effectiveness by finding successful cases in the second half of this year.
On this day, KT also introduced its self-developed RAG, 'K RAG,' a technology akin to a 'safety mechanism' that verifies the basis of AI responses before providing answers. KT highlighted as key differentiators: a RAG-specialized search model composed of embedding (converting text into vectors recognizable by AI) and reranking (reordering retrieved information to be closer to the correct answer) technologies; a modular search engine capable of industry-specific customization; and a maintenance system covering everything from data updates to redeployment.
Yoo Hee-hee, KT's Search AI Division lead, said, "Corporate data, policies, and terms of service are continuously updated, and user question patterns are becoming increasingly diverse. The core lies in establishing a 'virtuous cycle' that collects logs generated during operations and reflects user feedback."
<© STARNEWS. All rights reserved. No reproduction or redistribution allowed.>
*This content was translated by AI.
