Why Do Complicated AI Interfaces Become Simple?

Colin Dunn says complicated interfaces usually exist because the models are not good enough, and once the models get good enough the interfaces become very simple. A hard switch to an entirely new modality is too soon: the technology is not ready, and people are not ready.

People fall at different points along the adoption curve, from early adopters to the early majority, the late majority, and laggards. Customers right now are early adopters who see what is coming and want to be part of it. Tremendous economic and creative opportunity belongs to people who decide to jump in, and it is okay to stay skeptical and keep a way of working that already works. One inroad swaps searching for stock photos for generating a photo. Searching and prompting feel similar. Browsing a library for a photo that meets every criterion often feels non-creative, while generating a photo lets a person art direct and steer the model.

The interface, the buttons, and the mental model keep moving as the models get better. The job is to separate the part that will be obsolete from the load-bearing part: when a model improves dramatically, people will still want to think, operate, and manipulate imagery the same way. Because the customer is the professional designer, there is no need to sit on the frontier of every new thing that comes out.