Domestic AI Agent Trends (July 16, 2026 Issue)
Update Date: 2026/7/16
Executive Summary
As of July 15, 2026, the trend of AI agents in domestic companies has rapidly shifted from proof-of-concept experiments to company-wide deployment, implementation in core business operations, and the establishment of governance infrastructure. At companies like Resonac and TAPP, a democratization model where frontline employees create their own agents has taken root, resulting in high utilization rates and large-scale operational efficiency gains. Meanwhile, IBM Japan and Fujitsu are focusing on full-scale development and modernization support, incorporating company-specific development standards and expertise while enabling multiple agents to collaborate. In customer service, transcosmos, SoftBank, and Salesforce are deploying execution-oriented AI that goes beyond generating answers to handling procedures and business processing. Additionally, operational design—including safety, permissions, auditing, and cost management—has become a key theme, as seen in Toyota Tsusho's human-in-the-loop final verification, HENNGE's read-only connections, and GMO's specialized governance organizations.
*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 Resonac standardizes frontline-originated AI agents for company-wide deployment, reducing work by 470 million yen per month
📎 Source: Resonac Holdings Corporation Press Release
Resonac Holdings has built a unique model where AI agents created by frontline employees to address operational issues are refined for quality and standardized by headquarters before being deployed company-wide. Within about six months of introducing generative AI in October 2025, the utilization rate among the 2,405 target employees reached 95%. According to the company's calculations, this has resulted in a reduction of approximately 90,000 hours of work per month, equivalent to 470 million yen. For risk prediction activities, the company automated scoring and improvement feedback, reducing the time required by approximately 1/60th, and reduced the time for creating training materials for chemical regulations by approximately 1/5th. By expanding this to equipment installation, inventory management, and R&D support, the company has demonstrated an operational model that elevates frontline-driven results to corporate standards.
2️⃣ IBM Japan launches AI-driven development solution 'ALSEA'
📎 Source: IBM Japan Newsroom
IBM Japan launched its enterprise AI-driven development solution, 'AI Lifecycle Shared Engineering Artifacts (ALSEA),' on July 15. The solution systematizes company-specific development standards and expertise into a context that the 'IBM Bob' AI agent can utilize, while using a 'harness' to manage AI behavior and artifacts to ensure quality, consistency, and reproducibility. Complex development tasks are distributed to sub-agents, supporting requirements definition, design, and implementation while suppressing context bloat. Since April, the company has been conducting pre-verification for application to over 80 companies and is also verifying the feasibility of application in multiple application development areas with Mizuho Bank.
3️⃣ TAPP employees create approximately 500 business AI agents in 3 weeks
📎 Source: TAPP, Inc. Press Release
TAPP announced that it introduced the enterprise AI 'Gemini Enterprise' to all employees and achieved a 100% utilization rate in the three weeks from June 15 to July 5. The 194 target employees created a total of approximately 500 business AI agents, and AI responses reached 28,317. Based on the company's estimate that 15,800 effective responses saved 5 minutes each, this has streamlined approximately 1,317 hours of work. Prior to the introduction, an AI strategy unit directly under the CEO designed information security and permission management, establishing an operational foundation based on the premise of universal usage. While details regarding agent autonomy were not provided, this is a case of internal AI democratization where the frontline leads the increase in use cases.
4️⃣ Toyota Tsusho introduces 'TechTouch AI Hub' to expense settlement system
📎 Source: TechTouch, Inc. Press Release
Toyota Tsusho has introduced 'TechTouch AI Hub' to its existing business trip and expense reimbursement system. An AI agent with receipt OCR automatically checks dates, payees, and tax rates, as well as inconsistencies with the company's specific itemization requirements and dining rules, before the application is submitted, with final confirmation handled by a human. A key feature is the ability to retrofit AI support onto a web browser without modifying the existing system. The company aims to reduce application errors and rejections, streamlining the expense settlement process by approximately 73,000 times per year. There are plans to gradually expand this to approval tasks and other core systems, making it a case that balances automation and control based on a human-in-the-loop premise.
5️⃣ transcosmos integrates autonomous AI agents into its CX platform
📎 Source: transcosmos inc. Press Release
transcosmos has integrated autonomous AI agent functions into its CX platform 'trans-DX for Support,' enhancing it to 'trans-DX Plus for Support.' By combining technology from AI agent specialist vottia with the knowledge the company has accumulated through operations with approximately 3,500 companies, it links self-resolution of inquiries with human support. In one implementation case, switching a voice bot for repair reception to a voice AI agent increased the self-resolution rate from approximately 19% to approximately 52%. The platform features multi-LLM, encryption, guardrails, and transaction management with communication infrastructure, and the company plans to introduce it to 200 companies this fiscal year.
6️⃣ SoftBank and Sierra begin domestic sales of customer service AI agents
📎 Source: SoftBank Corp. Press Release
SoftBank entered into a strategic partnership with Sierra and began an initiative on July 14 to exclusively sell the company's conversational AI platform in Japan. Goal-oriented AI agents understand inquiry intent and autonomously execute not only answer generation but also various procedures and product return processing. Verification of effectiveness in LINEMO customer support showed that compared to existing services, the resolution rate improved from 83% to 97%, and customer satisfaction improved from 74% to 93%. Moving forward, the company is considering introduction to SoftBank, Y!mobile, and group companies, and will support Japanese companies with design, environment construction, implementation, and operation.
7️⃣ Salesforce announces autonomous AI teammate 'Agentforce Coworker'
📎 Source: Salesforce Japan Co., Ltd. Press Release
Salesforce Japan has announced its autonomous AI teammate, 'Agentforce Coworker.' Through Data 360, it understands corporate context such as deals, contracts, cases, and service history, and from a single conversation, it calls upon Agentforce's specialized agents, CRM actions, Flows, and external APIs to execute complex tasks. It continues to plan in the background even when the user is away from their desk, and can be used with the same context from Salesforce, Slack, Microsoft Teams, ChatGPT, and Claude. The design inherits existing permissions, business rules, and guardrails, balancing the convenience of execution-oriented AI with corporate control.
8️⃣ Fujitsu automates legacy modernization with multi-AI agents
📎 Source: Fujitsu Limited Press Release
Fujitsu has launched the "Fujitsu AI-Driven Modernization Service" in Japan. By combining Fujitsu Kozuchi, the "Takane" LLM, Claude, GPT, and others, a modernization-specific AI agent handles orchestration, parallel task execution, language conversion, verification, and improvement loops. While adopting a self-evolving configuration where multiple agents collaborate and learn, human intervention is included in final decision-making to manage quality and risk. By reflecting insights from thousands of projects into specialized agents, the company claims it can reduce legacy system rewrite and rehost processes by approximately 40%.
9️⃣ HENNGE provides an MCP server to connect security training data to AI agents
📎 Source: HENNGE K.K. Press Release
HENNGE has launched an MCP server for its targeted attack email training service, "HENNGE Tadrill." From MCP-compatible AI agents or AI assistants like Claude, users can directly retrieve training results, report emails, e-learning performance, and user attributes to perform departmental comparisons, identify high-risk users, analyze suspicious email trends, and create periodic reports through dialogue. The goal is to reduce manual work between traditional dashboards and spreadsheet/document creation tools. The MCP connection is read-only, the Tadrill usage fee is free, and the AI assistant is designed to be provided by the user company, clearly defining the boundary between agent integration and security.
🔟 GMO Internet Group establishes a specialized organization for AI agent promotion and governance
📎 Source: GMO Internet Group, Inc. Press Release
GMO Internet Group established the "Group AI Promotion Headquarters" on July 14 to advance the implementation and deployment of business AI agents across the group. The organization consists of four offices: the AI Platform Office, which handles API, MCP, and CLI connections and infrastructure development; the AI Agent Business Support Office, which handles business inventory and agent design; the AI Operations Office, which manages costs, contracts, and operations; and the AI Governance Office, which handles regulations, audits, and incident response. By managing everything from development to knowledge, tokens, costs, core system migration, and security in an integrated manner, they have launched an AI agent governance model premised on company-wide deployment.
Comprehensive Analysis
The characteristics observed from the topics on July 15, 2026, suggest that the axis of competition for AI agents has shifted from model performance itself to the ability to connect with corporate data, the execution power to complete tasks, and the standardization capability for company-wide reuse. Companies achieving particular success are balancing centralized control and decentralized emergence, where management leads in preparing infrastructure and rules while the front lines discover and develop use cases. Furthermore, with the spread of orchestration to oversee multiple specialized agents and connections via MCP, APIs, and existing systems, AI is evolving from a standalone dialogue tool into an entity that operates corporate systems. However, caution is required in simple comparisons, as the calculation conditions for time savings and resolution rates published by each company differ. Moving forward, the success of adoption will depend on operational design that incorporates performance measurement including quality, revenue, risk, and customer experience—rather than just usage frequency—along with authority management, audit trails, and human approval.
Future Points of Interest
Agent lifecycle management—who reviews, quality-assures, organizes duplicates, and promotes front-line-created AI agents to corporate standards—will become critical.
As AI agents move from response support to execution areas such as applications, returns, approvals, and system updates, the design of least privilege, audit trails, stop functions, and human approval points will become a competitive advantage.
With the expansion of connection destinations via MCP and APIs, boundary design that limits writing and external transmission while safely utilizing corporate data will determine the speed of adoption and security.
Companies that can track not only published time savings and resolution rates but also error rates, rework, customer satisfaction, revenue contribution, and operational costs using common metrics will improve the accuracy of their investment decisions.
As multi-agent systems advance, the completeness of the overarching infrastructure that assigns work to appropriate agents and safely recovers from failures will make more of a difference than the type of LLM adopted.
The presence of a specialized organization that centrally manages token costs, contracts, knowledge updates, and incident response while maintaining education and freedom to encourage universal usage will determine the speed of company-wide deployment.
