This is a continuation of “Where is all the software?”

The apps I use still look the same, feel the same, react the same, have the same speed, the same buttons, the same quirks. Is AI coding not good enough, or are we, as the users, not enough for AI?

Most software progress comes the moment you notice the product behaving differently in the wild and you take that human created mess and codify it into a new behavior, or squash it out of existence. Those learnings come from support threads, ops alerts, sales calls, onboarding. So most changes are driven by users.

AI scaffolds, debugs, refactors, writes tests. That’s why you feel faster, and are faster. But until market signals are produced at the same rate as code is created, the speed gains will stay on the coding side. The rate limiter at the moment is the market, we’re stuck with the same human‑paced feedback loop.

The thing that will flip this is a different type of user. If human users are replaced at scale by agent users acting on their behalf, the signals change faster. When the consumer is a bot, the discovery loop moves faster and we would be able to optimize for these usage patterns directly.

Which then opens the question, do we still optimize for human-navigable UIs, click flows, polished animations and that stuff, or do we focus on schema changes, API contracts, minimum latency? Should we even bother creating UIs for machines or will the API be the product?