renue Develops 'Agent Monitor' (Desktop Version) to Real-Time Evaluate and Improve Employee Prompts

renue (Headquarters: Minato-ku, Tokyo; CEO: Yusuke Yamamoto) has developed a desktop version of its service, 'Agent Monitor,' which visualizes AI agent usage. The application resides on each developer's macOS and evaluates the quality of instructions (prompts) written in Claude Code in real-time. It sends notifications with improvement tips for instructions lacking clear objectives or completion criteria. The AI automates the business improvement feedback that was previously handled by supervisors.

Executive Summary Effective management requires a consistent feedback loop. Observing subordinates' work, providing feedback, and helping them apply it to the next task is key to improving results. However, managers are often busy with routine tasks. Watching subordinates' AI usage and providing instant guidance is difficult. There is a lack of personnel on the ground to provide feedback.

renue has developed a tool where AI handles this feedback. It is the desktop version of 'Agent Monitor.' It resides on each developer's macOS and evaluates the quality of instructions written in Claude Code in real-time. For instructions that are not aligned with the objective, it returns improvement hints via macOS notifications. Because insights are delivered immediately after writing, subordinates can revise their instructions on the spot without asking a supervisor for help. Notifications are delivered silently and only when necessary, without interrupting workflow. The evaluation reads the last three interactions and does not penalize approvals or brief follow-up instructions.

In essence, AI handles feedback to subordinates in real-time when needed. Managers can enhance the quality of the team's AI utilization without sacrificing their own time. This is renue's vision of management in the AI era.

What is 'Agent Monitor'? 'Agent Monitor' is a service that visualizes the usage of AI agents such as Claude Code, Codex, and Cursor across an organization. Usage history from each terminal is automatically aggregated to a server, and usage status and prompt quality can be verified via an administrator dashboard. It also features real-time detection of dangerous operations and prompt quality evaluation. renue launched the service on March 19, 2026.

The desktop version expands this foundation to individual terminals, implementing the real-time feedback loop. It connects organizational visualization with on-the-spot individual guidance in a single product.

Premise of the Targeted Area Enterprise adoption of generative AI is moving from the introduction stage to the stabilization stage. Many workplaces have started using agents like Claude Code, Codex, and Cursor daily. The stage of installing tools is over; the question now is how to master them as an organization. In this phase, management is receiving new types of inquiries. Customers and internal departments are asking how to elevate the team's usage of AI integrated into their work.

Good or bad instructions cannot be measured by keywords or character counts. Even the same short instruction can be appropriate or inappropriate depending on the preceding context. Mechanical rules would penalize legitimate instructions. Even for humans, judging instructions requires experience, and such personnel are limited. It is not about how much AI is used, but how well it is used that determines results. The tools are there, but the gap in proficiency remains. The mechanism to bridge this gap and maintain improvement is still lacking in the field.

Goal renue aims to maintain an improvement cycle for AI usage through feedback on prompts without increasing management burden. We aim for a form where subordinates can improve their usage in daily tasks without supervisors having to check each instruction.

To this end, we have AI handle the improvement feedback. Instead of waiting for training or post-facto reviews, we provide feedback the moment the instruction is written. We do not demand too much at once, but rather convey the single most effective hint for that situation. It is not a lecture, but a suggestion on how to write more effectively next time.

If feedback cycles automatically, managers can spend time on their original decision-making tasks. Subordinates improve without relying on supervisors. Since notifications only appear when necessary, managers can focus on the key points. The goal is to disconnect the starting point of improvement from the supervisor's availability. By accumulating individual improvement, the quality of the entire team's AI utilization is raised.