Domestic AI Agent Trends (July 23, 2026 Issue)
Update Date: 2026/7/23
Executive Summary
In the domestic AI agent trends for July 22, 2026, advancements were made simultaneously in implementations that process tasks directly within business screens or existing SaaS, and in mechanisms for humans to verify and control those actions. Concre launched a construction-specialized agent capable of handling changes to estimates and budget data, while bellFace announced that Nakamura Tome Precision Industry reduced the time spent entering sales records by approximately 80% after introducing bellSalesAI. SOBA has automated pre- and post-sales recording and scheduling, and Nippon Life Insurance is advancing AI-driven efficiency in creating response records and streamlining management department verification. Jira and PROMPT MAFIA are progressing with delegation to multiple agents and the visualization of development processes. Including Box's announcement of guardrails, the axis of competition is shifting from answer generation to "business data updates," "cross-system execution," and "human approval and auditing."
Note: 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 Concre AI launches AI agent that directly modifies construction estimates and budgets through dialogue
📎 Source: Concre Inc. Press Release
Concre Inc. has added a conversational "Estimate Creation and Budget Management AI Agent" to its "Concre AI" for small and medium-sized construction companies. When instructed on the estimate or budget screen, it supports adding or moving line items, adjusting quantities, unit prices, selling prices, planned costs, and gross profit margins, as well as budget restructuring due to post-order specification changes or additional construction. AI modifications are displayed as unconfirmed drafts, allowing users to verify them before finalization, and it also supports cancellations and history checks. When combined with existing estimate AI agents, this creates a seamless data management flow from initial drafts based on drawings or past estimates to ordering, cost management, and billing, serving as an example of AI being embedded into construction business execution screens.
2️⃣ Nakamura Tome Precision Industry reduces sales record entry time by approximately 80% with bellSalesAI
📎 Source: bellFace Inc. Press Release
Machine tool manufacturer Nakamura Tome Precision Industry has introduced bellFace's Salesforce entry agent, "bellSalesAI," to its sales department. They have built a workflow that automatically extracts and structures key points from sales conversations, registers them according to Salesforce fields, and notifies Teams. According to bellFace's announcement, after the introduction, the time spent entering sales records was reduced from 30 minutes to 5 minutes per case, an approximately 80% reduction. Staff members who previously struggled to maintain consistent entry habits are now recording information, leading to improved information density and specific feedback from management. This is a case study in the domestic manufacturing industry demonstrating the value of AI agents through actual operational time savings and data consistency.
3️⃣ Ureru Net Kokokusha Group pre-releases "SOBA Sales AI" to cross-automate sales administration
📎 Source: Ureru Net Kokokusha Group Inc. Press Release
SOBA Project, a subsidiary of Ureru Net Kokokusha Group, has begun the pre-release of "SOBA Sales AI," which automates sales administration. It analyzes sales audio and email history to handle meeting minutes creation, next-action proposals, automatic updates of deal status, and the creation of reply drafts. It integrates with Google Calendar and Outlook to detect confirmed schedules and automates appointment registration and the setting of online meeting URLs. It proposes potential deals to approach next based on past order history and analyzes patterns of won and lost deals. It can be used solely within a browser and acts as a sales agent that performs repetitive administrative tasks before and after meetings across multiple tools, with plans to gradually expand the number of user companies.
4️⃣ Nippon Life Insurance introduces AI agent "UPWARD" for field sales to corporate sales divisions nationwide
📎 Source: UPWARD Inc. Press Release
Nippon Life Insurance has introduced UPWARD's AI agent for field sales, "UPWARD," to its corporate sales divisions in branches nationwide. To address the challenges of time-consuming recording of information obtained during customer interactions and the difficulty of sharing information due to variations in input content and methods, they are streamlining record creation and reinforcing management department verification tasks with AI. UPWARD is a service that provides automatic recording of customer touchpoints through advanced geofencing and AI functions specialized for field sales. Nippon Life Insurance plans to combine customization tailored to business logic with consulting to reduce the burden on the front lines, improve sales productivity, and enhance the use of information for proposal activities. This adoption by a major insurance company across its nationwide branches is attracting attention.
5️⃣ Atlassian centralizes development work management for humans and multiple AI agents in Jira
📎 Source: Atlassian K.K. Press Release
Atlassian has added a new set of features to Jira to manage development across humans and AI agents. Jira Planner generates technical specifications from the codebase and Jira/Confluence history, Agents in Jira assigns work items directly to Claude Code, Cursor, GitHub Copilot, etc., and Agent sessions in Jira visualizes bottlenecks, pending reviews, and completion status on a single screen. By providing requirements, architecture, decisions, and dependencies as context via Teamwork Graph, internal benchmarks showed a 44% improvement in answer accuracy and a 48% reduction in token usage compared to agents without context. This represents a shift toward an AI-native development foundation that maintains human accountability while visualizing autonomy.
6️⃣ Remy Consulting begins offering "PROMPT MAFIA," where role-based AIs collaborate on development
📎 Source: Remy Consulting LLC Press Release
Remy Consulting has begun offering "PROMPT MAFIA," a monthly service where multiple AI agents collaborate to advance software development. The requester handles requirements and final decisions, while agents specialized in roles such as advisor, implementation, vulnerability inspection, testing, and release judgment work in a chain. Models and roles are organized for each project, and PM reports from requirement organization to implementation results are automatically generated. While they promote 24-hour development and the acceleration of prototyping and revisions, they standardize consultant support because quality depends on requirements and instructions. This is an example of an implementation-type multi-agent system where humans retain evaluation and final decision-making rather than full automation.
7️⃣ Box announces guardrails and auditing features that also cover external AI agents
📎 Source: Box Japan Inc. Press Release
Box, Inc. has announced security and governance features to control AI agents inside and outside of Box that handle enterprise content. It provides operation and access restrictions based on confidentiality levels, prompt injection detection before reaching the model, and scoped permissions for external agents via Box MCP. It also supports behavior monitoring, audit trails, retention of session logs, and human approval before critical operations. They plan to expand these controls to Claude, ChatGPT, Gemini, and others, with a phased rollout to Enterprise Advanced plan users over the coming months. This move indicates that the axis of competition for agent adoption is shifting beyond functionality to include permissions, auditing, and the ability to halt operations.
Comprehensive Analysis
The key feature observed from the topics on July 22, 2026, is that we have entered a phase where competition is shifting from 'what to answer' to 'which business data can be updated, with what permissions, and to what extent.' Concur and SOBA directly modify business objects such as quotes, budgets, schedules, and project information, while bellSalesAI and UPWARD connect unstructured information generated in the field to CRM systems. In Jira and PROMPT MAFIA, operations have materialized where humans retain requirements, evaluation, and final judgment while distributing tasks to multiple agents. Meanwhile, the access control, prompt injection detection, audit trails, and human approval demonstrated by Box become indispensable as the scope of execution expands. Moving forward, it will be important to design not only model performance but also the ability to confirm change differentials, reversibility, minimization of privileges, log retention, and actual operational KPIs as an integrated whole. In particular, services that can continuously disclose field metrics—such as the reduction from 30 minutes to 5 minutes announced by bellFace—will be more likely to progress from PoC to company-wide deployment.
Points to Watch in the Future
A point of interest is whether Concur's design for change proposal confirmation and cancellation will be applied to subsequent tasks with greater impact, such as ordering and coordination with partner companies.
For bellSalesAI, SOBA, and UPWARD, future comparison axes will include not only input time but also actual operational KPIs such as input rates, project update rates, proposal preparation time, and closing rates.
In Jira, the scope of information when passing organizational context to external coding agents, permission management, and cross-model cost management are critical.
In multi-agent development like PROMPT MAFIA, the focus is on the rationale for each agent's judgment, the reproducibility of vulnerability testing, and the design of final human approval.
The specific timing of the rollout of Box's guardrails to the domestic environment and the functional constraints for each external agent connected via MCP will influence adoption decisions.
