When the model IS the product, all UX is RLHF. Every edit, correction, and click becomes a training signal. Product designer’s new craft: building for learnability, not just usability.

Each era of computing had its own learning loops. PC era: manual loops e.g. bug reports. Web era: programmatic loops e.g. clicks. AI era: self-learning loops e.g. tabs, accepts/rejects, drafts.

In AI products, the interface is the feedback surface. Every Accept, Edit, or Delete teaches the model something. So the product design question is no longer just “is it easy to use?" it’s also “can the system learn from use?”