Most AI products feel broken on first use. Not because the AI is bad.

Most AI products feel broken on first use. Not because the AI is bad. Because no one had time to design them.

There is a documented three-month gap in AI agent development where UX simply does not happen. Engineers ship. Designers clean up afterward. By then the architecture has decided everything.

I see it with clients. The demo lands. The handoff fails. Real users do not know what to type. They do not know when to wait. They do not know what went wrong. Nobody designed those moments because they were not on the roadmap.

The fix is not more designers on the team. It is design earlier in the process. Before the first endpoint. Before "we'll polish this in v2."

If you are building an AI product: sketch the failure states before you sketch the success states. What does the interface look like when the model is slow? When it is wrong? When it is confused? Those screens matter more than the happy path.