Domestic AI Agent Trends (August 12, 2026 Issue)
Update Date: 2026/8/12
Executive Summary
As of August 11, 2026, the use of AI agents in Japan has progressed from simple conversational support to integration with actual business operations, including internal information retrieval, document generation, municipal guidance, agricultural machinery, marketing, e-commerce operations, and security control. ChatSense directly references internal documents on Box to integrate everything from answering questions to generating slides, while ENSOR and ECPRO BRAIN build business cycles from analysis to execution based on organizational knowledge and product data. Additionally, Shizuoka City is conducting trials with mobile AI avatars for administrative guidance, and Iwata City is advancing demonstrations connecting AI judgment with fruit sorting machines. On the other hand, as AI autonomy increases, access rights, approval history, audit trails, and human verification are becoming critical, with 'AI agent-specific privilege management,' represented by Saviynt's Zuma, emerging as a new foundational area.
*The created article content was input into Gamma to automatically generate slides. Please take a look here if you find slides easier to view.
1️⃣ ChatSense: Integrating everything from questions to slide generation using internal documents on Box
📎 Source: Knowledge Sense Inc. Press Release
Knowledge Sense has added Box integration to the Notebook AI feature of its corporate generative AI agent, 'ChatSense.' Users can directly import multiple PDFs and Office documents from Box without saving them to a device, and proceed from asking questions about the document content to generating slides in the same workflow. The company reports a product capability that can generate slides in about 2 minutes and a track record of over 500 corporate implementations for ChatSense as a whole. This announcement is a functional expansion of an existing product rather than an introduction to new customers. Internal document governance, such as consistency with Box access rights, the scope of imported documents, and human verification of generated content, will be the focus of actual operations.
2️⃣ Shizuoka City Shimizu Ward Office demonstrates window guidance using a 'walking AI avatar'
📎 Source: First Launch Inc. Press Release
A government office guidance demonstration combining First Launch's AI avatar 'AICO' and the autonomous robot 'Lanky Mini' began at the Shizuoka City Shimizu Ward Office on August 10. The kiosk type supports over 13 languages, and the robot type initially supports 4 languages; they approach visitors stopping near the entrance to offer proactive guidance for counters and procedures. Until the end of February 2027, the usage rates and guidance effectiveness of the screen-based and robot types will be compared, and students from the Shizuoka Institute of Science and Technology will participate in knowledge registration, response confirmation, and improvement proposals. While the demonstration results will be used for planning the new Shimizu government building, this does not guarantee full-scale implementation.
3️⃣ Iwata City launches a collaborative project connecting agriculture-specialized AI agents with fruit sorting machines
📎 Source: ReGACY Innovation Group Inc. Press Release
ReGACY Innovation Group announced that it has begun supporting five collaborative projects under Iwata City's 'Iwa-Sta Accompaniment Support Program.' Among these, 'Kyuri Tomato Nasubi' (Cucumber Tomato Eggplant), which develops agriculture-specialized AI agents, and Smart Agriculture Iwata are working on the development and introduction of a wedge-type paprika sorting machine equipped with AI automatic identification functions. The project aims to improve the efficiency and quality of the sorting process, and a kickoff event was held on July 22. It should be viewed as a project working on the development and introduction of a sorting machine that connects on-site machinery with AI-based judgment, rather than a track record of full-scale implementation of a completed machine.
4️⃣ REHATCH 'ENSOR' adds approval and change history to organizational shared memory
📎 Source: REHATCH Inc. Press Release
REHATCH has performed a major update to 'Brain,' the learning foundation of its marketing AI OS, 'ENSOR.' It now allows users to view and add brand information understood by the AI on the screen and supports the registration of brand fonts. Version 2.0 introduces approval steps and change history when sharing personal memory with a team, and version 3.0 allows the AI to automatically reference necessary content from accumulated information. The company claims that it processes analysis, appeal design, copy creation, and design generation in a series, reducing the time from analysis to production by up to 1/10th, with AI editing taking 3 minutes for revisions and 10 minutes for submission completion; however, these are vendor-claimed values, not third-party verified measured values.
5️⃣ ECPRO BRAIN proposes and executes improvement measures based on product and SKU-specific data
📎 Source: Proteinum Inc. Press Release
Proteinum has added performance management and AI analysis functions by product and SKU to its e-commerce-specialized AI agent, 'ECPRO BRAIN.' It can check sales, sales volume, and trends before and after implementing measures across Rakuten Ichiba, Amazon, Yahoo! Shopping, and some proprietary sites in a unified format. The AI analyzes issues and growth potential for each product to present improvement proposals, and existing functions also support execution and scheduling such as product name changes, coupon settings, and product CSV uploads. This is an example of a business-specialized agent that continuously cycles through the PDCA of e-commerce operations by connecting measure planning, task creation, reflection on malls, performance confirmation, and the next improvement proposal on the same foundation.
6️⃣ Saviynt announces 'Zuma' to control AI agent permissions at runtime
📎 Source: Saviynt Inc. Press Release
Saviynt has announced and made immediately available 'Zuma,' a security foundation for managing AI agents and non-human identities. 'Zuma Insights' continuously detects agents, LLMs, and AI applications within an organization, while 'Zuma Access' narrows down permitted actions based on intent, context, permissions, and risk at runtime. 'Zuma Governance' manages owners, lifecycles, access reviews, and audit trails. While details on implementation cases in domestic companies or support systems for Japan were not provided this time, the announcement indicates the formation of a market that controls AI agent permissions as a management target separate from human IDs.
Comprehensive Analysis
The characteristics observed from the topics of August 11, 2026, were that the value axis of AI agents is shifting from 'high-performance answers' to 'understanding, judging, executing, and connecting the results to the next action using business data.' In particular, connections with existing systems and physical spaces, such as Box, e-commerce malls, fruit sorting machines, and government office robots, are increasing, and AI is beginning to be incorporated as part of practical business processes. At the same time, mechanisms to control information that AI can reference and operations it can execute, such as approval and change history of shared memory and runtime access control, are becoming central to product value. In the future, it will be difficult to differentiate based on generation accuracy or speed alone, and it is believed that the 'business foundation of AI agents,' including connectivity with business systems, privilege management, auditability, and operational improvement, will become the axis of competition.
Future Points of Interest
As AI agents shift from 'AI that proposes' to 'AI that actually operates,' design for control—including minimization of execution authority, approval flows, and operation logs—is likely to become a more important indicator for adoption decisions than response accuracy.
For services like ChatSense and ENSOR that continuously utilize internal documents and organizational knowledge, memory governance—specifically 'whose information can be shared with AI, by whom, and to what extent'—will become a greater differentiator than search performance.
The ECPRO BRAIN case study indicates a direction where AI handles not only analysis but also product information changes and policy implementation; in the future, the 'ratio of tasks completed by AI' may become a more important evaluation metric than the 'proposal adoption rate.'
In local government and agricultural sectors, AI is moving beyond web screens into the physical world through robots and industrial machinery, and it will be interesting to see if cases of AI agent and robotics integration increase toward 2027.
As shown by Saviynt's Zuma, if the concept of managing enterprise AI agents as 'non-human identities' distinct from humans spreads, the IAM market itself may be reorganized with AI as a prerequisite.
Since many of these are currently in the proof-of-concept stage or based on vendor-claimed effects, it will be important to see how much real-world operational data—such as man-hour reduction rates, error rates, correction rates, and continuous usage rates—is disclosed in the future, rather than just the number of companies adopting them or generation speeds.
