Mitate turns plain-language requests into designed systems

Mitate's approach starts from a deliberately different entry point. Many ways of building a system ask the user to first decide what to make — a reservation sheet, a lending log, an inventory list — and that entry point favors people who can translate their problem into a system category. R2-san argues that someone who cannot make that translation gets stuck on the first step, and that even when AI is available, if it simply picks the nearest template and applies it, the result may not match the actual problem. The same label, such as equipment management, can require different things depending on why the user is stuck.

The design AI therefore checks three things instead: the purpose (what would need to be known for the problem to stop), the situation (who records what, and when), and what should be protected (who may be trusted with what). When a point that touches purpose or information handling has not been decided, it is returned as a question or left unset so it can be asked for later, while choices the user need not specify, such as field ordering, are left to the AI. Mitate also does not treat its existing building blocks as the ceiling on what a user may ask for; a necessary condition is something to be formally designed, though conditions it cannot handle today are declined and added later as product development.

In the live test, the request named neither an app type nor any fields, yet the result included a response record, an automatic notice to all members, and an organizer-sent reminder. The designed screen also showed a proposal it had not adopted: deciding who may read the operation log, which the request never mentioned. That is a concrete illustration of the stated aim — that the person who can describe the problem should be able to take part in designing its solution.