AI Learns Rules and Know-how for Operations... KT Establishes 'Agentic Data Flow'

Reducing the Complexity of Collaboration Among AI Agents
Expanding to 'AI for Everyone' and Beyond... Enhancing the Ecosystem

KT has introduced a new work operation model in which humans and artificial intelligence (AI) agents collaborate. The AI agents interpret data based on the knowledge accumulated by KT, ensuring that the data remains up to date and provides consistent results in line with company standards. The entire process runs seamlessly from end to end.

Sangbong Lee from KT AX Future Technology Institute AX Data Lab is explaining the 'Agentic Data Flow' on the 3rd at Club806 in Jongno-gu, Seoul. Photo by Gyojo Noh

Sangbong Lee, Head of the AX Data Lab at KT AX Future Technology Institute, held a briefing on September 3 at Club 806 in Jongno-gu, Seoul, and stated, "We have established a company-wide data utilization system called 'Agentic Data Flow,' which enables AI agents to independently understand corporate data and leverage it for operations."

The Agentic Data Flow is designed to go through a series of steps—K Feature Map, K Meta, KT Ontology, and K Evaluation. One of its key strengths is resolving the complexity of cooperation between agents to spread AI throughout the organization. The 'K Feature Map' serves as a common data asset that standardizes and integrates company-wide KT data, including customer enrollment, usage patterns, products and pricing plans, and marketing. Currently, more than 1,600 types of datasets that have passed security reviews are provided to ensure privacy and data protection. This has significantly reduced the time required for data exploration and processing.

The data governance platform 'K Meta' provides integrated management of a dataset's location, structure, relationships, quality, and security policies. This enables the AI agents to adhere to privacy and confidentiality policies, and prevent data misuse and security risks. 'KT Ontology,' which structures internal terminology, work manuals, and knowledge into a unified semantic framework, transforms human experience and know-how into digital assets and lays the groundwork for work automation. Lee explained that applying KT Ontology improved accuracy by 62.4% compared to previous systems.

Additionally, KT operates 'K Evaluation,' which assesses the entire work process—including planning and tool invocation. This is a quality evaluation system designed to enhance the reliability and operational stability of the AI agents. KT has established a virtuous cycle for improvement by performing comprehensive assessments based on an in-house benchmark tailored for the Korean language and local industry environment, analyzing root causes, and implementing solutions whenever issues arise.

KT is applying the Agentic Data Flow to a variety of services, including MyK, Sales Agent, and Small Business Agent, continuously improving its completeness. The company plans to expand this initiative to a work automation agent called 'K-Claw' (tentative name). By linking it with the 'Everyone's AI' service and broadening the application of agents to other sectors, KT is advancing its agentic AI ecosystem. Lee said, "Leveraging the assets and experience we have accumulated in the most complex data environments in Korea, we aim to present a new work operation model where people and agents work together."

This content was produced with the assistance of AI translation services.