Domestic AI Agent Trends (August 14, 2026 Issue)

Update Date: 2026/8/14

Executive Summary
The trends in domestic AI agents as of August 13, 2026, show that AI agents have expanded from "conversational software" into execution platforms capable of writing to external systems, code conversion and verification, commercial transactions, incident response, accounting, and contract operations. Representative examples include CRM registration and notifications by AI AgentPlus Pro, autonomous verification by Smartrans, EC connectivity by DG Agentic One, Teams integration by Incident Lake, and post-payment processing by TOKIUM. At the same time, developments such as Hubble's isolated execution environment, Comix's multi-agent operations, untactit's approval and auditing, and participation in the Osaka Prefecture consortium indicate that the axis of competition is shifting beyond model performance to include authorization, verification, operational design, and governance.

*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 ChatPlus "AI AgentPlus Pro" automates everything from customer support to CRM registration, notifications, and email sending

📎 Source: ChatPlus Inc. Press Release
ChatPlus launched "AI AgentPlus Pro" on August 13, which automates everything from customer support to subsequent tasks. The AI understands and judges inquiry content and customer intent, executing not only responses but also conversation summaries, CRM registration in Salesforce or kintone, staff notifications via Slack, and email sending. External system integration can be built without code, aiming to reduce routine input, transcription, and internal sharing. Since it involves writing and sending, it is important to define the granularity of permissions, approval procedures, stop/recovery protocols in case of misexecution, audit logs, and handling of personal information for each business flow upon implementation.

2️⃣ Daiwa Institute of Research "Smartrans" applies AI-based code conversion and consistency verification to large-scale migrations

📎 Source: Daiwa Institute of Research Ltd. Press Release
Daiwa Institute of Research has expanded the quality control functions of its legacy migration tool "Smartrans," which utilizes generative AI, and has begun full-scale application to large-scale system migration projects. It optimizes code structure and notation to match the target language characteristics and architecture, combining this with quality checks to improve readability and source structure uniformity. In the patented current-to-new consistency verification, AI agents autonomously repeat code conversion and consistency checks between old and new systems. The design also organizes post-migration specifications and configuration information in a format easily referenced by generative AI, supporting maintenance, impact analysis, and continuous improvement.

3️⃣ Digital Garage "DG Agentic One" integrates EC front-end for AI with MCP

📎 Source: Digital Garage, Inc. Press Release
Digital Garage and DG Business Technology have launched the integrated platform for EC operators, "DG Agentic One." It combines DataFeed, which structures product information for AI; GEO, which supports recommendations in AI search; MCP, which handles product/transaction search and cart generation by supporting UCP/ACP; and Commerce, which adds a front-end for AI agents. It is a platform that connects external AI agent recommendations to purchases without requiring large-scale modifications to existing EC sites, with a goal of supporting 3 trillion yen in transaction volume by 2030. The payment function, Payment, is under development, and specification confirmation regarding identity verification, user approval, spending limits, and cancellation procedures will be the focus moving forward.

4️⃣ SIGQ "Incident Lake" structures incident response within Microsoft Teams

📎 Source: SIGQ Inc. Press Release
SIGQ has released the Microsoft Teams app for its incident management AI agent, "Incident Lake," on the Marketplace. Users can create an incident with "go" on Teams, link related threads with "link," supplement posts made before the bot joined with "fill," and check progress with "ask." After integration, it captures discussions in real-time, updates structured timelines, severity, and status, and generates response time detection and post-incident report drafts on the web side. While it connects daily chat to records and decision-making support, usage requires an Incident Lake account, and corporate-side Teams app review and information access scope settings are essential.

5️⃣ Sumitomo Mitsui Card x TOKIUM develop corporate card linked with accounting AI agent

📎 Source: Sumitomo Mitsui Card Company, Limited Press Release
Sumitomo Mitsui Card and TOKIUM have formed a comprehensive business alliance in the corporate payment sector and are developing the "TOKIUM Business Card" as the first step. Aiming for a December 2026 launch, the accounting AI agent will retrieve card statements after payment, reflect them in the management screen, and automatically fill in account titles. It can issue alerts or send back non-compliant usage, and manage spending limits and usage periods per card. By connecting payments, vouchers, expense settlements, and accounting processes, they aim to reduce accounting burdens and strengthen internal controls. As it is currently pre-launch, it is necessary to confirm the basis for automatic rejection, exception handling, and the scope of human approval in the official specifications.

6️⃣ Hubble "Contract Flow Agent" expands cross-contract investigation and isolated execution platform

📎 Source: Hubble Inc. Press Release
Hubble has expanded the foundation of its contract AI agent, "Contract Flow Agent," and launched autonomous search and investigation functions that explore and analyze contract data across the board. The AI does not fix search conditions at once, but repeats additional searches and filtering based on results, traversing contract text, comments, change history, ledger information, and similar documents. At the same time, it built "AI Workbench," which executes programs in an isolated cloud environment for each request to perform document creation/correction, calculation, large-scale analysis, and deliverable verification. Some features, such as autonomous reflection and saving to files, are future development concepts, and it is necessary to evaluate currently available features separately from those to be added sequentially.

7️⃣ Comix releases sales operation model for autonomous AI agents

📎 Source: Comix Co., Ltd. Press Release
Comix has released a design model for autonomous AI agents used in its own management and sales operations. Once a request is made via LINE, it references calendars, past history, and customer information to select necessary business meetings, perform company analysis, and create agendas, proposal materials, and follow-up content; according to the company, it prepared materials for 37 meetings in about 15 minutes. After the meeting, it links the minutes to content analysis, customer information organization, deal scoring, follow-up policies, and the creation of proposal documents of about 30 pages. In the Claude Code environment, 55 agents and 15 scheduled executions are incorporated, with humans making final decisions, and maintenance and security processes such as API key leak detection are also in operation.

8️⃣ PoliPoli participates in the Osaka Prefecture Administrative AI Agent Consortium

📎 Source: PoliPoli Inc. Press Release
PoliPoli has announced its participation in the "Osaka Prefecture Administrative AI Agent Consortium," established by Osaka Prefecture. This consortium is a framework for investigating and analyzing the current state of the AI agent field and concretizing the potential for utilization in local governments through demonstration projects. PoliPoli will bring its expertise in policy co-creation and public-private partnership to consider applications for "Local Government Management AX," which aims to improve resident services and enable the creation and execution of policies even with limited personnel. This announcement is a declaration of participation rather than a product introduction or the start of an individual demonstration; future points to confirm include the administrative procedures involved, data usage, staff approval points, accountability, and impact assessments on residents.

9️⃣ untactit provides a Japanese-language foundation for controlling AI agent instructions and memory

📎 Source: untactit Press Release
untactit has launched the Japanese version of its "Agent Asset Control Plane," which manages the skills, instructions, and memory referenced by AI agents across the board. It inventories assets from multiple environments such as Claude, ChatGPT, Gemini, Cursor, and Copilot, detects duplicates, and distributes approved versions to all terminals. It separates personal and company-wide assets, features four roles (administrator, approver, editor, and viewer), append-only audit logs, local drift detection, and distribution withdrawal. As it is pre-official launch with a small number of customers and explicitly states that certain certifications have not yet been obtained, it is necessary to verify certification status, data storage locations, connection permissions, and recovery procedures before implementation.

Comprehensive Analysis

The features observed from the topics on August 13, 2026, indicate that the domestic AI agent market is shifting from a stage of increasing standalone chat screens to a stage of combining reading/writing of existing systems with operational decision-making. The scope is expanding to include CRM registration and notifications in customer support, code conversion and verification in development, search and cart generation in e-commerce, statement retrieval and rejection in accounting, and cross-sectional search and deliverable verification in legal affairs. On the other hand, as autonomy increases, erroneous transmissions, registrations, purchases, and judgments lead to real operational losses. Therefore, it is necessary to incorporate approval points, least privilege, audit logs, isolated execution, rollback, and exception handling not only as product features but also as part of operational design. In addition, it is important to disclose quantitative KPIs such as time saved, error execution rates, and human intervention rates, and to evaluate safety and effectiveness using the same indicator system. Future competitiveness will be determined not just by "what can be done," but by the ability to clarify "whose authority is used, what evidence is left, and where a human can stop the process."

Future Points of Interest

• For write-capable agents like AI AgentPlus Pro or TOKIUM, the feasibility of implementation depends on separating permissions for reading and execution, and determining where to place approvals before transmission, registration, or rejection.

• DG Agentic One has presented support for MCP, UCP, and ACP along with an AI-oriented e-commerce front end, but Payment is still under development. The focus is on the standardization of identity verification, user approval, and cancellation/refunds during payment.

• For code conversion agents like Smartrans, it is more important to be able to leave audit trails for current-to-new consistency verification, quality gates, and responsibility boundaries than to focus on generation speed.

• In cases like Incident Lake and Hubble, the design of embedding agents into existing work environments such as Teams or contract SaaS is progressing. The challenge lies in balancing on-site adoption with secure data boundaries.

• In the Osaka Prefecture Consortium, as it moves from a declaration of participation to actual demonstrations of administrative procedures, attention will be focused on the handling of resident data, explainability, final approval by staff, and the disclosure of effectiveness measurements.