Gemba AI Development Project

Last updated: October 11, 2026

Creating a Place Where Humans and AI Collaborate

The age of industrial AI is dawning. Yet much of what advanced solution providers offer is aimed at automating what happens inside and around the factory. We don't believe that fully automated, machine-only AI is the right shape for industrial AI. Large enterprises aside, most on-site workplaces around the world, including in Japan, are sole proprietorships or small and medium-sized businesses that employ many people. Replacing all of them with AI is simply not realistic, and we believe that what our society truly needs is industrial AI that creates a place where people and AI collaborate on-site. Our motto is not to let AI be monopolized by large corporations or a handful of advanced nations.

Here is an example of a use case for Gemba AI, currently under development, set in a manufacturing workplace ("gemba" means "on-site," that is, the actual place where the work happens, in Japanese).

• [Understand] AI deployed on-site observes the work, converses with workers, and, where needed, references operational documents and equipment IoT data to understand the entire workplace (devices support voice conversation rather than text chat; IoT data is planned as an optional feature for the beta).
• [Document] Through its understanding of the workplace (AI-driven process mining), the AI documents on-site know-how, including business processes and decision-making rules (understanding workers' tasks and skills, and converting tacit knowledge into explicit knowledge, while also grasping the overall on-site operations).
• [Support] Based on the documented business processes and decision rules, the AI supports a range of on-site tasks, from routine operations to exception handling, anomaly response, and on-site training.
• [Report] The AI turns the status of on-site work and operations into reports and dashboards.
• [Accumulate Experience] As it supports workers and gains experience with the work, the AI updates its documentation of on-site know-how (including updates to the AI agent's skills and tools, as well as integration with process mining, BPMS, and BRMS tools).
• [Automate] The AI identifies tasks that can be automated with apps or robots, provides or recommends apps and robot skills (in coordination with physical AI) to replace that work, and takes the lead in driving AX/DX transformation on-site.

This is only part of the functionality, and even so, turning just these capabilities into a robust system that reliably handles a wide variety of requirements is an extremely difficult undertaking.

Beyond manufacturing, there are many other types of on-site workplaces in our society, including construction, retail, repair, restaurants, and tourist destinations. Everyday living spaces count too, and looking further into the future, there will be places where people work and live on lunar bases, space stations, and spacecraft. We are advancing the development of a system platform to create "a place where humans and AI collaborate" across these many settings in society.

What Real-World Skills and Real-World Tools Should We Give Real-World AI Agents?

Today's AI agents are advancing primarily in text-centric domains, built around a core mechanism known as the harness, which requires skills, tools, and sub-agents to be designed (by a human designer) to fit the target application. Agent Skills are also becoming increasingly standardized.

Gemba AI is a real-world AI agent, and it inherits this current AI agent framework. That makes the design of skills and tools critical. However, the skills and tools needed to operate AI agents safely and reliably in the real world do not yet exist. Moreover, since we are not aiming for full on-site automation but rather a balanced handling of people, AI, and the environment together, figuring out how to structure these skills and tools is a very difficult problem. We must consider how to handle images, IoT sensor data, conversations with people, and the many kinds of information present in the real three-dimensional world, in order to support human-AI collaboration, that is, the work itself, in a wide range of situations. We call these real-world skills and real-world tools, and we are currently advancing research and development using a variety of scenarios.

Connecting to Digital Twins and Physical AI

Gemba AI is not merely digitizing the workplace, nor merely automating it. While providing a human-centered foundation for on-site AX/DX, it also aims to provide tools for capturing on-site information that help transition the workplace toward digital twins (digitalizing the workplace mainly through simulation) and physical AI (automating the workplace mainly through robot deployment). Introducing digital twins and physical AI requires collecting and processing highly accurate, comprehensive on-site information. While creating a place where humans and AI collaborate, Gemba AI provides an opportunity to gather this kind of on-site three-dimensional position data and human work data (physical data). (This remains at the planning stage for beta development.)

Industrial AI Aimed at Supporting On-Site Operations

In this era where AI has become commonplace, we continue to take on the daily challenge of creating, starting with manufacturing workplaces, an industrial AI that supports on-site operations centered on human-AI collaboration. This is neither mere on-site digitalization or a digital twin, nor physical AI aimed at automating work.

We are targeting a release of this product in the first half of 2027, aiming for a leading-edge solution that can support the operations of sole proprietors and small and medium-sized businesses, even if only in a small way.