Domestic AI Agent Trends (August 10, 2026 Issue)

Update Date: 2026/8/10

Executive Summary
From August 7 to August 9, 2026, domestic AI agents steadily progressed from simple generative AI response support to "task-execution types" that continuously perform multiple steps. At DotAI, 12 agents handle tasks ranging from collecting and updating municipal systems to development and information dissemination, while Money Forward has automated consolidated financial statement reviews, and the drone sector has automated flight log evaluation and improvement guidance. Meanwhile, with the expansion of corporate adoption, control and safety have become key themes, as seen with Headwaters entering the 'Agentic Security' market, which includes authority management and human approval. Furthermore, movements like Speeda AI Agent and WonderAgent, which combine existing data infrastructure, external systems, and human expert support with AI, are spreading. AI agents are shifting from a stage of competing on autonomy itself to a stage of competing on implementation into business workflows and role division with humans.

*The content of the created article was input into Gamma to automatically generate slides. If you find slides easier to view, please take a look here.

1️⃣ DotAI 'Estatix Child-rearing Support', sharing business operations among 12 AI agents

📎 Source: .AI TIMES 'A service that compares child-rearing subsidies for 135 municipalities nationwide, automatically operated by 12 AI agents'
.AI TIMES has released a case study of operating the child-rearing subsidy comparison service 'Estatix Child-rearing Support' with 12 AI agents. Targeting approximately 135 municipalities and 2,500 to 3,000 systems in Tokyo, Kanagawa, Chiba, and Saitama, agents such as planner, collector, updater, and validation share tasks including information gathering, differential updates, and recalculation verification. Inquiry responses, SNS posts, article writing, front-end implementation, and DB design are also automated, with agents reportedly executing most of the 160 commits in the last 30 days. While weekly policy proposals and UI mocks are also automatically created, the final execution decision is handled by the operators.

2️⃣ Drone flight support 'AI Coach', automating everything from flight log scoring to improvement feedback

📎 Source: Japan Drone Business Support Association Press Release
The Japan Drone Business Support Association has launched an 'AI Coach' in its flight support app 'dronebiz.app' and released it as a limited beta version. It reads flight logs from DJI aircraft such as the Phantom 4 series and Mavic series, and scores six perspectives—smoothness of stick operation, hovering stability, altitude maintenance, vertical movement, straight-line movement, and circle/figure-eight patterns—on a 100-point scale. It also supports 3D replays and deduction evaluation based on the national second-class practical examination, passing the results to the conversational AI 'aotoriAI' to present improvements in Japanese. Although not an autonomous planning type, it is an execution-type workflow that continuously processes everything from log import to evaluation and feedback, demonstrating the deployment of AI in skills training.

3️⃣ Headwaters enters the 'Agentic Security' market to control corporate AI agent operations

📎 Source: Headwaters Co., Ltd. Official Announcement
Headwaters has announced its entry into the 'Agentic Security' market, which safely supports the production use of AI agents. Its core architecture, 'Agentic Security Harness,' is a concept that links AI agents, existing security products, company-specific business rules, and human approval. It continuously tracks IDs, permissions, data connections, and execution history, making threat detection, impact analysis, response, and verification a continuous operational cycle, while automating low-risk processing and returning important decisions to human approval. At this stage, it is not an announcement of introduction to specific customers, but a stage of advancing technical verification and creating use cases with client companies, a move to commercialize control design itself.

4️⃣ uSonar and Uzabase partner for BtoB sales support utilizing Speeda AI Agent

📎 Source: uSonar Inc. Press Release
uSonar and Uzabase have announced a business partnership to support BtoB sales and marketing in an integrated manner, from targeting to policy execution and effectiveness verification. uSonar handles customer data integration, deduplication, trend identification, and potential customer identification through uSonar and PlanSonar, which are centered on the corporate database 'LBC'. Starting from the created target list, Uzabase is responsible for policy implementation preparation, approach design, and post-execution progress and effectiveness verification, proposing the combined use of Speeda AI Agent, Speeda MCP linkage, and human support through consulting as needed. This is not the start of a fully autonomous service, but the construction of a collaborative system that connects data infrastructure, AI agents, and human expert support.

5️⃣ Money Forward provides a review agent that automatically checks for anomalies in consolidated financial statements

📎 Source: Money Forward, Inc. Press Release
Money Forward has started providing the AI agent 'Consolidated Financial Statement Review Agent' for 'Money Forward Cloud Consolidated Accounting'. It automatically checks consolidated packages and consolidated financial statements (balance sheets and income statements) for balance consistency, theoretical value consistency, and anomaly candidates based on comparisons with the previous period and the same period of the previous year. It classifies results into categories such as 'Needs Confirmation,' 'Caution,' and 'No Issues,' and if an anomaly is detected, it lists the relevant account items and estimated causes as 'AI-based correction hints,' designed to allow staff to focus on significant discrepancies. Future plans include sequential support for segment-based income statements, cash flow statements, customization of check rules, and anomaly detection through statistical and time-series analysis.

6️⃣ Wonderful Fly begins free verification support for corporate AI agent 'WonderAgent'

📎 Source: Wonderful Fly Inc. Press Release
Wonderful Fly has launched the 'WonderAgent AI Introduction Support Campaign,' which allows companies to verify AI agents in actual business for free. WonderAgent is a corporate platform that continuously executes multiple processes necessary to achieve goals, such as information gathering and organization, analysis, judgment, document creation, and system operation, based on user instructions and business flows. It combines standard functions and company-specific functions as 'AI Skills,' and also supports linkage with external systems and the use of multiple generative AI models through Amazon Bedrock. In addition to providing WonderAgent for free, it is an initiative to make it easier to verify business suitability and introduction effects before full-scale implementation by providing technical support for initial settings and usage methods.

7️⃣ Pencil begins PoC where 5 AI agents operate an internal event

📎 Source: Pencil Co., Ltd. Official Release
Pencil has launched an internal event called 'AISCREAM,' hosted by five AI agents called 'φ’s (Phis),' as a demonstration experiment to explore how organizations where humans and AI grow together should be. Agents with different personalities and roles pose questions to employees, encourage challenges, and support actions, while promoting the provision and use of functions utilizing G-DX through the independently developed AI platform 'miniΦ.' This event is the first step in the 'φ’s Co-evolution Spiral OS Concept,' where humans and AI create change while influencing each other. It is not a business automation product for the general public, but an internal experiment that has finished the planning stage and moved to the PoC stage to verify the impact of AI on organizational behavior and learning.

Comprehensive Review

The key characteristics observed from the topics between August 7 and August 9, 2026, indicate that the domestic AI agent market has moved past the stage of aiming for 'all-purpose agents that autonomously do everything' and has entered a phase of clearly breaking down specific tasks and integrating data, AI, and human approval into practical operations. Implementation is particularly advanced in areas where input data and evaluation criteria are relatively clear, such as information updates, financial statement reviews, sales support, and skills assessment. On the other hand, as the number of agents and the volume of automated processing increase, the importance of operational control—such as access management, log auditing, halting operations during malfunctions, and defining areas of responsibility—also grows. It is believed that what will determine future competitiveness is not just model performance, but the level of sophistication of the 'agent operational infrastructure' that can securely connect company-specific data and business rules, and design the boundaries between areas where AI should make decisions and areas where humans should make the final judgment.

Points to watch in the future

• The evaluation criteria for AI agents are shifting from 'how far they have achieved autonomy' to 'how stably they can reproduce business results.' In addition to processing volume and man-hour reduction, the disclosure of practical operational KPIs, including quality and error rates, is likely to become important.

• As configurations where multiple agents share tasks become widespread, it is expected that Agentic Security—which integrates the management of authorization, execution logs, anomaly detection, and escalation to humans—will become a prerequisite for corporate adoption, even more so than individual AI performance.

• Agentification is progressing in areas where correct answers and judgment criteria are easy to define, such as accounting, sales, municipal information, and skills training. The next focus is on how much reproducibility can be ensured in areas involving ambiguous judgment, such as planning, customer support, and organizational management.

• As shown by cases like Speeda AI Agent and WonderAgent, if configurations that connect company-specific data, MCP, external systems, and multiple LLMs spread, 'what it can be safely connected to' will become a more important differentiator in product selection than the AI model itself.

• Pencil's PoC is an attempt to use AI not just for business substitution, but also for behavioral change and learning promotion among employees. Moving forward, it is worth paying attention to a new evaluation axis: 'AI agents that enhance organizational capabilities,' which cannot be measured by productivity improvement alone.