Examining the Effectiveness of PC Operation Agent Technology for Japanese Businesses

TDSE Inc., based in Shinjuku, Tokyo, along with Mitsubishi Research Institute (MRI), has successfully validated its PC operation agent technology, which allows AI to control PCs based on human instructions, for Japanese business contexts. With this achievement, the companies have planned to initiate consulting and implementation support to apply this technology to client operations.

Background and Motivation

The PC operation agent, also known as a GUI agent, is a technology where AI utilizes a mouse and keyboard to execute commands based on human input. A variety of software incorporating this technology, both commercial and open-source, have emerged, showcasing a significant potential for business use. However, there has been a scarcity of validation data regarding its efficacy within Japanese business practices, leading to concerns about its quality.

In a collaborative effort, TDSE's Data Technology Lab and MRI's AI Consulting Department progressed with the validation of the PC operation agent’s capability, affirming its effectiveness in business environments where Japanese is the primary language.

Figure 1: Image representing the operation of the PC operation agent

Validation Methodology and Results

To verify the quality of the PC operation agent technology in a Japanese business setting, TDSE and MRI prepared a validation environment along with datasets modeled after common business activities, such as system configurations and administrative tasks. They tested the performance of various mainstream commercial and open-source software solutions.

The results showed that, depending on the application areas and system layouts, these agents could reach a level where practical implementation could be considered viable. In scenarios where functions can be inferred from the interface and button names, the agents are capable of executing accurate commands. Notably, the open-source PC operation agents demonstrated satisfactory performance in specific business workflows, paving the way for potential applications involving highly confidential information when supported with proper environments.

Consequently, this technology allows for the automation of diverse tasks on systems without requiring specialized configurations, distinguishing it from existing technologies like RPA (Robotic Process Automation).

Additionally, the design and construction of benchmarks tailored for business applications have yielded valuable insights regarding the conditions of execution environments, task structuring, evaluation methods, and reproducibility, all of which are critical for conducting appropriate evaluations. This accumulated knowledge stands to benefit practical applications considerably.

For further details, please refer to our technical blog:

- An Era Where AI Operates Screens on Its Own: Building Unique Benchmarks and Exploring Practical Applications of GUI Agents (TDSE, February 10, 2026)
- AI Blog: Episode 12 - Validation of GUI Agents (PC Operation Agents) and Its Practical Applicability (Mitsubishi Research Institute, February 10, 2026)

Figure 2: Evaluation datasets and results

Future Plans

Moving forward, TDSE and MRI will deepen their collaboration to advance the validation of cutting-edge technologies and their social implementation. The insights obtained regarding the utilization and evaluation of PC operation agents during this undertaking will be employed in MRI’s generative AI consulting, contributing to the improvement of client productivity and reduction of operational burdens. Likewise, TDSE will harness the technical knowledge gained from this validation to create and provide AI agent services tailored to client needs, bolstering business digital transformation and competitiveness.

Company Overviews

TDSE Inc.

1. Name: TDSE Inc.
2. Location: 27th Floor, Tokyo Opera City Tower, 3-20-2 Nishishinjuku, Shinjuku-ku, Tokyo
3. CEO: Naoki Higashikata
4. Business: Solution provision utilizing big data and AI, including AI products (including AI modules)
5. Capital: 833.18 million JPY
6. Established: October 17, 2013
7. Division: Data Technology Lab (Promoting research and validation of data analysis and AI technology)

Mitsubishi Research Institute

1. Name: Mitsubishi Research Institute
2. Location: 2-10-3 Nagatacho, Chiyoda-ku, Tokyo
3. CEO: Kenji Yabuta
4. Business: Think tank and consulting services, IT services
5. Capital: 6.3624 billion JPY
6. Established: May 1970
7. Division: AI Consulting Department (Supporting business transformation for clients using AI and data)