Domestic AI Agent Trends (August 8, 2026 Issue)
Update Date: 2026/8/8
Executive Summary
In the domestic AI agent market as of August 7, 2026, the shift from departmental-level proof-of-concepts to full-scale integration into enterprise-wide data infrastructure and specialized business operations has become clear. Japan Post Insurance is advancing company-wide deployment through a comprehensive agreement with Salesforce, while Loglass has announced a 75-80% reduction in man-hours for management accounting tasks. Vertical-specialized agents that handle continuous processing by referencing multiple documents and external data are expanding across freee, RAKUS, LegalOn, and advertising operations firms. The axis of future competition is shifting beyond high autonomy to operational design, including data connectivity, permission management, evidence presentation, and human-in-the-loop approval.
*I input the created article content into Gamma to automatically generate slides. If you find slides easier to view, please take a look here.
1️⃣ Japan Post Insurance signs comprehensive AI agreement with Salesforce, accelerating company-wide use of Agentforce
📎 Source: Salesforce Japan Co., Ltd. Press Release
Japan Post Insurance has concluded a comprehensive AI agreement with Salesforce, which the company announced as a first for a domestic corporation. They are already using 'Agentforce Financial Services' in contact centers and parts of sales, and 'Agentforce Marketing' in marketing, and plan to expand the scope of application company-wide. They plan to connect data dispersed within the company using 'Data 360' and 'MuleSoft' and deploy it across a wide range of touchpoints, including customer support and content delivery. While the agreement, which relaxes usage volume constraints, will accelerate adoption, company-wide access permissions and data governance will become critical.
2️⃣ Loglass officially launches individual company-specific AI agent business specializing in management and finance
📎 Source: Loglass Inc. Press Release
Loglass has officially launched an 'AI Solutions Business' that develops individual company-specific AI agents specializing in the management and finance domain, supporting everything from PoC to full-scale implementation. It structures management context across DWH, financial figures, chats, and meeting minutes, handling data collection, integration, forecasting, analysis, document generation, notifications, and chat responses. MIXI announced that it reduced monthly report creation and review time from 96 hours to 24 hours (approx. 75%) and performance aggregation from 60-80 hours to 12 hours (approx. 80%). Future focus will be on reproducibility for other companies and the quality evaluation of deliverables.
3️⃣ freee announces 4 types of AI agents connecting tax accountant operations and a development platform
📎 Source: freee K.K. Press Release
freee has announced the rollout of 'freee Client Management | AI Agent' for tax accountant offices and the no-code development platform 'freee Agent Hub.' The former consists of four types: bookkeeping, rule maintenance, monthly checks, and tax return checks, which cross-reference vouchers, statements, BS, PL, and tax documents, returning low-confidence processing to human review. Agent Hub allows for the isolation of authentication information and files for each client, enabling the creation, sharing, and scheduled execution of proprietary agents. freee plans to provide the tax return check agent for free and the monthly check agent for free for a limited time starting August 28, along with the official launch of 'freee Agent Hub,' with the paid version of 'freee Client Management | AI Agent' scheduled for release on September 15; this announcement precedes these start dates.
4️⃣ NTT DOCOMO supports careers with a multi-AI agent system for 25,000 employees
📎 Source: NTT DOCOMO, INC. Press Release
NTT DOCOMO announced that its career development support using multi-AI agents on the HR AI platform 'Job-Voyage' won the Excellence Award in the HR System category at the 11th HR Technology Awards. Four agents—grasping the current situation, considering target images, proposing actions, and overseeing them—reference work history, skills, and training records to individually configure the flow of dialogue. They provide a voice dialogue environment to approximately 25,000 group employees, and more than half of those who received learning suggestions from the AI moved on to actual learning activities. This is a domestic case study demonstrating behavioral change through large-scale operation.
5️⃣ Kirin operates AI agents resident in research workflows in some research departments
📎 Source: Kirin Holdings Company, Limited Press Release
Kirin Holdings and GenerativeX announced that they have built an 'AI-native research environment' that embeds AI agents into research activities themselves, and have been operating it in some of Kirin's research departments since 2026. Rather than researchers calling on AI whenever needed, it is constantly embedded in the flow of hypothesis formation, information exploration, brainstorming, documentation, and knowledge sharing, accumulating exploration history and discussions as organizational knowledge. Future plans include expanding functions to include understanding research themes and expertise, detecting signs of issues, proposing hypotheses, and facilitating collaboration between researchers. Quantitative results have not been disclosed, and contribution to research quality is the next evaluation axis.
6️⃣ Ricoh Japan and LangGenius support Osaka Prefecture's administrative AI agent demonstration with Dify
📎 Source: Ricoh Japan Corp. News Release
Ricoh Japan is participating in the 'Osaka Prefecture Administrative AI Agent Consortium' promoted by Osaka Prefecture, and in collaboration with LangGenius, will provide Dify licenses and utilization support. Dify allows for the no-code development of AI agents and complex AI workflows, and because it supports self-hosting and various AI models including domestic LLMs, it is intended for implementation according to the security and network requirements of each local government. Through demonstrations with Osaka Prefecture and municipalities within the prefecture, they aim to streamline administrative and counter services, achieve high-level automation, and develop reproducible implementation models and guidelines. This announcement is a framework for participation and support, and the target operations and evaluation KPIs have not yet been announced.
7️⃣ RAKUS automates everything from voucher collection to confirmation of Electronic Book Preservation Act requirements
📎 Source: RAKUS Co., Ltd. News Release
RAKUS has begun offering the 'Voucher Acquisition AI Agent' for 'Rakuraku Electronic Storage,' which continuously processes everything from the collection of invoices received via email to registration and confirmation of Electronic Book Preservation Act requirements. In the current function, when an email is forwarded to a dedicated address, it automates the extraction of attachments, system registration, extraction of password candidates from the email body, and decompression of ZIP files, notifying the person in charge if it cannot be unlocked. A function to periodically patrol mailboxes to actively acquire vouchers is scheduled for release within 2026; currently, email forwarding is the starting point. It is a practical agent that combines routine processing with human intervention for exceptions.
8️⃣ Hakuhodo DY ONE expands 'Advertising Flow' to ChatGPT ad operations
📎 Source: Hakuhodo DY ONE Inc. Press Release
Hakuhodo DY ONE and Hakuhodo Technologies have added features for ChatGPT advertising to their advertising operation AI agent, "Advertising Flow." The AI analyzes delivery performance daily, organizes budget progress, projected landing figures, and signs of early termination for each project to notify Slack, while also proposing bidding actions based on the analysis results. The goal is to reflect insights gained from domestic pilot operations into the features, supporting a stable operational structure from the initial stages of ad delivery. However, the extent to which recommendations are automatically reflected in ad accounts is not explicitly stated, and it is necessary to confirm the process where humans judge and apply them.
9️⃣ E-Logi, Inc. begins beta provision of an agent for suggesting negative keywords for Google Ads
📎 Source: E-Logi, Inc. Press Release
E-Logi has begun providing a beta version of a "Negative Keyword Suggestion Agent" for search ads within "AD EBiS Campaign Manager." Once product and target information are registered, the system automatically retrieves Google Ads search query reports, and the AI suggests negative keyword candidates, methods such as exact match or phrase match, and the reasoning behind the judgment based on user operations. The design requires the person in charge to review and approve the suggestions, following a Human-in-the-loop model that leaves the final decision affecting ad delivery to humans. Currently, it only covers Google Ads; the beta version is free, and the official version is planned to be provided as a paid option.
🔟 LegalOn implements AI contract review that references master agreements and original contracts
📎 Source: LegalOn Technologies, Inc. Press Release
LegalOn Technologies has enabled AI contract review that references master agreements, original contracts, and memoranda of understanding within "LegalOn," which is equipped with a legal-specialized AI agent. When a user selects a saved document or adds a file, the AI checks and points out issues in the target contract using the content of related contracts as a basis for judgment, and automatically links the used documents as related files. This feature reduces the manual work of opening multiple documents for cross-referencing and expands reviews based on contractual relationships. On the other hand, it is not a mechanism where the AI automatically searches for related documents; the selection of reference targets and the final legal judgment remain the responsibility of the user.
Comprehensive Analysis
The characteristics observed from the topics on August 7, 2026, suggest that the domestic AI agent market is moving from the "standalone chat" stage to an implementation phase of "reading business data, connecting multiple processes, and returning only exceptions to humans." At Japan Post Insurance and Osaka Prefecture, data integration infrastructure and operational models for company-wide and cross-organizational use were established first, while for Loglass, freee, RAKUS, and LegalOn, the context of specialized operations—management, accounting, vouchers, and contracts—has become their competitive edge. Furthermore, NTT Docomo has demonstrated behavioral change, and Loglass has shown labor reduction, with performance evaluation beginning to shift from the number of PoCs to business KPIs. Meanwhile, in advertising and accounting, designs that leave final approval to humans remain mainstream. Future differentiation will arise not from the level of autonomy, but from the design capability to incorporate authority separation, evidence presentation, audit logs, exception handling, and quality measurement into business workflows.
Future Points of Interest
Regarding the company-wide deployment at Japan Post Insurance, attention will be focused on how cross-departmental access rights, auditing, and usage management are designed in addition to data integration using Data 360 and MuleSoft.
For freee, RAKUS, and E-Logi, it will become easier to evaluate practical investment effects if metrics such as the percentage of cases returned for human review, error rates, and correction time are disclosed.
Regarding the labor reduction and behavioral change demonstrated by Loglass and NTT Docomo, it is important to see if the effects can be maintained during long-term operation and if they can be replicated in other departments or companies.
In the administrative AI agent demonstration in Osaka Prefecture, the focus is on whether the target municipalities, specific tasks, evaluation KPIs, and division of responsibilities can be clarified and translated into a procurable operational model.
In ad operations and contract reviews, the authority boundary—whether the AI only makes suggestions or also executes system operations—and the auditability to track the basis for judgments will determine the feasibility of adoption.
