Classmethod obtains patent for technology that reconstructs individual knowledge and decision patterns as inheritable AI agents; ghoost records 92.3% Pee approval rate in UHB pilot

Classmethod Inc. (Headquarters: Minato-ku, Tokyo; President and Representative Director: Satoshi Yokota; hereafter “Classmethod”) obtained a patent on April 30, 2026, for technology that reconstructs the knowledge and decision-making patterns of individuals within an organization as inheritable AI agents. Its generative AI service “ghoost,” which uses this patented technology, recorded a Pee approval rate of 92.3% in an early deployment at Hokkaido Cultural Broadcasting Co., Ltd., showing that it has reached the practical-use stage. The individual decision-making know-how accumulated inside organizations has long been lost whenever employees retire or transfer. Although initiatives such as reducing dependency on specific individuals and knowledge management have been discussed for many years, methods for inheriting decision-making itself had not previously been systematized. This patented technology structures the inheritance of organizational knowledge and directly addresses a core issue in human capital management. The Pee approval rate indicates the percentage of cases in which third parties other than the target person, such as colleagues or stakeholders, approved ghoost’s generated responses as “like the person.” The fact that this figure is recognized by people around the individual rather than by the individual themself demonstrates the technology’s ability to inherit organizational knowledge. ghoost is an AI agent service that reproduces a person’s decision-making patterns based on their thoughts, thinking, knowledge, and memories. Through automated interviews and integrated processing, it can generate a first version of an AI agent in about one to two hours per person, enabling organization-wide deployment in a short period. The break-even point for deployment is designed to be extremely low. For example, in an organization with 300 employees, the monthly fee can be recovered if each employee reduces meeting time by slightly less than 10 minutes per week. This lowers the adoption barrier while contributing to productivity improvements across the organization. Hokkaido Cultural Broadcasting (UHB) began early deployment in March 2026. Because of the nature of program production, the broadcasting industry relies heavily on individual judgment, experience, and sensibility. At UHB, the long-standing issue was that there was no way to pass on “that senior colleague’s axis of judgment” to younger employees. This deployment was carried out as the organization’s first structured attempt to verify the inheritability of such person-dependent know-how. The proof-of-concept results covered 16 target individuals and 162 sessions, achieving a Pee approval rate of 92.3%. Ayuki Sugimoto of UHB’s Media Division DX Promotion Center commented that, at first, the team viewed intellectual property creation as systematizing the “thinking methods of excellent employees.” However, through the verification process, they came to realize that the accumulation of small daily judgments made by employees, not only those of a few outstanding people, is itself organizational know-how and can become an asset. This gave them confidence that judgments previously scattered across the workplace and dependent on individuals can be transformed into assets that the whole organization can inherit. The patent is titled “AI Agent Provision System, AI Agent Provision Method, and AI Agent Provision Program.” Its registration number is Patent No. 7857639. It was registered on April 30, 2026, and issued on May 13, 2026. The application number is Japanese Patent Application No. 2025-135907, with parent application No. 2025-134990 and related divisional application No. 2026-001423. It includes 18 claims, and the inventors are Satoshi Yokota, Hirokazu Tsuguma, Tsutomu Tateno, and Ryota Moriwaki. The core technology of the patent defines a method for generating AI agents as “role models” optimized for specific positions and responsibilities by integrating feature values from multiple AI agents. This makes it possible to build decision-making models that do not depend on a specific individual and can be inherited by the organization. The main technologies included in the patent are role model generation through the integration of multiple AI agents, automated interviews and personality model extraction workflows, mechanisms for cross-organizational reference and operation of role models, and accuracy improvement through continuous learning and updating. Classmethod also announced a ghoost release webinar. Hokkaido Cultural Broadcasting is continuing to consider future deployment and will speak at the webinar “Turning Tacit Knowledge into Organizational Power with AI: Official Launch of ghoost,” scheduled for June 4, 2026. Classmethod is a technology partner that supports corporate digital transformation, focusing on cloud-native technology areas including Amazon Web Services (AWS), data analytics, mobile, IoT, AI, and machine learning. In AWS support, the company has continuously been certified as a top-tier partner since 2015 and has won Japan’s “AWS Consulting Partner of the Year” award five times. In 2022, it won the global “SI Partner of the Year” award and was a finalist again in 2023, making it one of the world’s leading AWS partners in both name and substance. The company has supported about 5,600 companies and has helped manage and build more than 40,000 AWS accounts in total. It also emphasizes a culture of technical communication by engineers, publishing more than 60,000 technical articles on its owned media DevelopersIO and operating Zenn, a knowledge-sharing platform for engineers, thereby contributing to the growth of the technical community.