Policy
A Cabinet decision now asks for a division of roles between people and AI
Japan's Basic Plan for Artificial Intelligence, adopted 23 December 2025, states that the obstacle is implementation rather than research. Its fourth pillar is effectively a design brief.
Government documents are usually written carefully. That makes the parts written less carefully worth reading.
Chapter 1 of the Basic Plan for Artificial Intelligence, a Cabinet decision dated 23 December 2025, states that AI cannot be said to be actively used in Japanese life or work, and frames the principal obstacle as a shortfall in implementation rather than in research and development.
The plan's stated goal is for Japan to become the country where AI is easiest to develop and use.
The diagnosis of an implementation gap
This matches the statistics. The 2026 White Paper on Information and Communications puts generative AI usage at 86.4% of Japanese companies, firms reporting no organisational effort at 27.0% against 1.4% in the United States, and firms that have redesigned a process around AI at between 6.0% and 11.9%.
Adopted, not implemented. The government and the data are saying the same thing.
For companies, what matters is that this diagnosis determines where policy support goes. Frame the problem as research and support flows to universities and institutes. Frame it as implementation and support flows to the companies doing the adopting.
That is visible already. The IT introduction subsidy was renamed the Digitalisation and AI Introduction Subsidy for 2026, with a ceiling of ¥45 million and a subsidy rate reaching three quarters.
About the fourth pillar
The plan sets out several pillars. The fourth asks for a clear division of roles between people and AI.
That phrasing is unusual in a technology policy document. It is not about model capability, compute infrastructure or data availability. It is about allocation. Who does what.
It also happens to be a definition of design work.
- Which judgements stay with people
- Which tasks belong to the system
- Where the handoff happens
- Who notices when it goes wrong
Deciding those four is what we call process design. The same content is now written into a Cabinet decision as a national objective.
By putting the division of roles into a pillar, the government has formally reclassified AI adoption as an organisational design problem rather than a procurement one.
What a company should take from it
Three practical implications.
Support for implementation is increasing. The subsidy framework has shifted toward the adopting side. Names and eligibility change annually, so it is worth tracking the next application window.
Documenting the division of roles makes approvals easier. In a subsidy application and in an internal approval alike, a document that states what AI handles and what people handle gets through. A document that omits it stalls in the risk discussion.
The implementation gap is not only your problem. 27.0% means a large number of companies are stuck in the same place. Getting past it is itself a relative advantage.
What the plan does not say
The plan sets a direction. It does not describe a method, which is correct, because that is for each company to decide.
But if the diagnosis is right, what is needed is not new technology. It is the work of getting existing technology into the operating process. That work has a name, needs an owner, and takes time.
The Cabinet decision says that capability is nationally scarce. Which is another way of saying that few organisations currently have it.