ENEOS Materials built over 1,000 custom GPTs and put ChatGPT Enterprise in front of every employee

The interface is the deployment: Japanese prompts instead of tooling

The claim that stands out in this announcement is organizational, not technical. ENEOS Materials reports over 90% of employees using ChatGPT Enterprise at least weekly and more than 1,000 custom GPTs created across the company. That volume only happens when the way you interact with the system is the same as the way you already work — in this case, describing a task in Japanese and getting a result back.

Taku Ichibayashi of the R&D Department frames adoption as gated by two front-end concerns rather than raw capability: security for proprietary information and output accuracy. The team's stated approach — 'master the technology ourselves first, and then explore its potential' — describes an interface being learned by hand across departments before it was scaled, not a top-down platform mandate.

A non-coder shipping a tool is the real signal

The HR example is the concrete proof that the surface matters. Marie Takeda built an internal data-aggregation tool despite having, in her words, no prior coding experience, and reports roughly 90% less time spent on aggregation. A separate HR custom GPT compresses training analysis from one to two hours of manual work to about 20 seconds.

It was my first time trying coding, but with ChatGPT, I was able to create the tool myself without any coding knowledge.Montana Labs

For applied teams, this is the frontend story in miniature: the person who understands the training-feedback problem builds the solution directly, without a handoff to an engineering queue. The natural-language surface collapses the distance between someone who has a problem and someone who can build against it.

Deep research and custom GPTs as domain-specific frontends

The other use cases show the same pattern applied to specialized work. At the Hungary plant, Kenichi Sakemi's team uses deep research to search Hungarian-language sources and return precise Japanese, cutting investigations that once took months to tens of minutes and chemical-engineering calculations from half a day to minutes. The interface here doubles as a translation and search layer over local materials the team could not otherwise process quickly.

The Engineering department's plant-design custom GPT takes structured inputs — fluid type, flow rate, pipe diameter, pressure loss, material requirements — and returns optimized specifications against company standards, with material corrosion checks Sakemi says now take seconds. Each custom GPT is effectively a purpose-built front end wrapping the company's own standards, and the count of 1,000-plus suggests employees are building these interfaces for themselves rather than waiting for central IT.

The implication: adoption scaled because the barrier was language, not skill

What ENEOS Materials demonstrates is that in a manufacturing workforce facing labor shortages, the constraint on AI value was never model intelligence — it was who could operate the tool. By making the operating instruction 'describe what you need in Japanese,' the company turned domain experts in HR, process engineering, and plant design into builders without retraining them as programmers.

Sakura's stated ambition — to control shop-floor equipment 'in everyday language, guiding and optimizing production as easily as we interact with ChatGPT' — extends the same premise from office work to machinery. The bet is that the conversational surface, not a specialized control panel, becomes the default interface to complex systems. The 1,000 custom GPTs are the early evidence that when the interface matches how people already communicate, adoption self-propagates rather than requiring a rollout to be pushed.