Domestic AI Agent Trends (May 28, 2026 Issue)
Update Date: 2026/5/28
Executive Summary
The domestic AI agent trends as of May 27, 2026, indicate that the utilization of AI agents by domestic companies is shifting from the PoC stage to integration into actual business operations. In insurance assessment, Agentic RAG has achieved high accuracy and labor reduction in payment eligibility decisions and benefit amount calculations. Movements to automate business processes while supporting human judgment are spreading in accounting/tax, banking, and contact centers. Additionally, services emphasizing task completion and safe execution control, such as autoregressive agents, AI secretaries, and browser-operating agents, have emerged, signaling that the domestic AI agent market is entering a full-scale phase centered on specialized operations, finance, and support for small and medium-sized enterprises.
*The content of the created article was input into Gamma to automatically generate slides. Please take a look here if you find slides easier to view.
1️⃣ Finatext/Nowcast/Medicare Life, PoC for benefit payment assessment using Agentic RAG
📎 Source: Finatext Holdings Inc. Press Release
Finatext and Nowcast announced the results of a PoC conducted at Medicare Life using generative AI for benefit payment assessment operations. In Phase 2, AI was tasked with everything from reading claim documents to determining payment eligibility and calculating benefit amounts, adopting autonomous exploration of policy terms and manuals via Agentic RAG, utilization of past cases through long-term memory, and the selective use of multimodal models. Concrete results were demonstrated, including an assessment accuracy rate of over 90%, AI-OCR accuracy of approximately 90%, a reduction in input work hours by about 1/3, and a reduction in assessment work hours by about 40%, marking an important case of applying agents to specialized insurance assessment tasks.
2️⃣ TKC announces "FX Agent" and "OMS Agent" for accounting and tax
📎 Source: TKC Corporation Press Release
TKC announced the "FX Agent" to be integrated into the cloud accounting system for small and medium-sized enterprises "FX Cloud Series," and the "OMS Agent" to be integrated into the business management system for TKC members "OMS Cloud." Some functions of the OMS Agent will be provided in advance in June 2026, with a prototype to be released in July and sequential provision starting from the end of September. Because they handle highly confidential data such as finance, payroll, My Number, invoices, and tax information, they emphasize Governance by Design, Platform-Native AI, and Human in the Loop. The design is characterized by supporting periodic audits, financial analysis, draft responses to inquiries, and progress management, rather than AI replacing the judgment of tax accountants.
3️⃣ GMO Aozora Net Bank announces AI agentization in 3 areas and Agentic API concept
📎 Source: GMO Aozora Net Bank Press Release
GMO Aozora Net Bank announced a medium- to long-term strategy to simultaneously AI-agentize three areas: "AI bank for customers," "AI transformation of the bank itself," and "AI bank for AI." Personalized banking by a dedicated AI agent for each customer is scheduled to begin in November 2026, and for internal operations, the bank has set a plan to replace 2,800 tasks with 1,100 AI agents by fiscal year 2028. Furthermore, it indicated a policy to build an "Agentic API" by March 2027, where AI autonomously selects and executes financial functions, and to proceed with support for public MCP.
4️⃣ Quantum Research Institute begins offering the "OWL" completion-type autoregressive agent
📎 Source: Quantum Research Institute Inc. Press Release
Quantum Research Institute released "OWL," a general-purpose autoregressive agent that repeats planning, execution, verification, and correction until a task is completed. With strengths in software development and stock monitoring, it highlights a design that breaks down user instructions into tasks and loops until a deliverable is created while detecting errors and missing information. It is an example of a domestic startup-type agent aimed at "completing" processes that previously required human supervision, such as requirements definition, implementation, verification, improvement, and report creation, symbolizing the shift in the axis of competition from dialogue quality to completion rate.
5️⃣ Puente begins offering "A-ko-san," an AI secretary dedicated to the president
📎 Source: Puente Inc. Press Release
Puente has begun offering "A-ko-san," an AI secretary service dedicated to presidents of one-person companies and small corporations. Centered on the Claude API, it operates on the company's "Puente Platform," accepting business instructions, progress checks, approvals, and notification summaries from the president via natural language dialogue, and relaying them to backend business systems or MCP server groups for automatic execution. With corporate pricing starting from 5.4 million yen for initial costs and 300,000 yen per month, it is also characterized by handling server operations and API management in bulk. It is attracting attention as a model for small businesses to build business operations with AI employees.
6️⃣ Knowledge Sense's "Cowork" strengthens risk judgment accuracy for browser-operating agents
📎 Source: Knowledge Sense Inc. Press Release
Knowledge Sense has improved the AI risk judgment accuracy in "Cowork," the browser operation function of its corporate AI agent "ChatSense." Cowork is an autonomous AI agent that can automatically execute browser operations such as searching, form entry, and data collection, and process information across multiple web services. In this improvement, it more accurately distinguishes between final operations that are difficult to undo, such as purchasing, deleting, and sending, and preparatory operations like "add to cart" or "deselect," requesting human confirmation only when necessary. This is important as a reinforcement of guardrails that balances autonomous execution safety with operational continuity.
7️⃣ Advanced Media adds real-time support function for multiple AI agents to AmiVoice
📎 Source: Advanced Media, Inc. Press Release
Advanced Media has added an "AI Assistant Function," where multiple AI agents provide real-time support to operators, to its AI speech recognition solution for contact centers, "AmiVoice Communication Suite 4.6." The AI agents infer based on the content of the call and not only present necessary information and support content but also execute tasks such as launching applications depending on the situation. This expands from the conventional use of generative AI centered on post-call summarization to judgment support during calls and execution support for subsequent tasks, serving as an implementation example that supports the entire process of contact center operations.
Comprehensive Analysis
A key highlight from the trends on May 27, 2026, was that AI agents are beginning to be designed not merely as chatbots or summarization tools, but as 'parts of business processes' that handle business decision-making, information retrieval, task execution, and progress management. Especially in fields requiring high accuracy and accountability, such as finance, insurance, and accounting, there was a clear emphasis on Human-in-the-Loop, guardrails, and Governance by Design, favoring the reinforcement of human professional judgment over full automation. On the other hand, initiatives like GMO Aozora Net Bank's Agentic API concept and end-to-end, resident-type agents like 'Fukuro' and 'A-ko' demonstrated the next stage where AI executes tasks across systems.
Points to watch in the future
AI agents are starting to deliver results in specialized tasks such as insurance assessment and accounting/tax affairs; moving forward, the explainability of the basis for decisions, rather than just accuracy, will become a competitive advantage.
GMO Aozora Net Bank's Agentic API concept is designed on the premise that AI autonomously selects and executes financial functions, making the progress of MCP (Model Context Protocol) support critical.
With the emergence of AI secretaries and end-to-end agents, the possibility for small businesses to possess operational foundations that compensate for labor shortages is expanding, with pricing models and implementation effectiveness becoming the focus.
In browser operations and contact center support, the convenience of autonomous execution and the risk of operational errors are two sides of the same coin, and the skill in designing confirmation processes will determine whether or not they are adopted.
In the domestic market, implementation capabilities—including industry-specific data, integration with existing systems, operational responsibility, and human approval—will become more of a differentiator than the performance differences of general-purpose AI.
