Crit on an AI conversion feature found the real design problem: a 5-minute wait. On progress, exits, completion sounds, and refining instead of restarting.
Nobody puts a waiting screen in their portfolio. This week’s crit made a case that somebody should. The work under review: an AI feature that turns uploaded paperwork into working templates, the kind of conversion that used to cost an afternoon of manual rebuilding. The catch sits in the middle: the conversion can run 5 minutes. And for those 5 minutes, the design question stops being what the feature does and becomes what the customer does. The wait is the interface.
The room treated it like a full surface, and the feedback stacked fast. Show estimated time, because an unlabeled spinner reads as a hang by minute 2. Offer a sound on completion, so the customer can safely leave the tab the way a kitchen timer buys you the rest of the house. Let them cancel. Give the wait a shape with a step indicator. And design the return: when they come back, the result and its controls should sit exactly where the attention lands, and the room had opinions about even the discard control’s position.
My favorite question came late: if the output is bad, can the customer refine it instead of throwing it away and re-uploading? Right now the answer is start over, and the room called out what that costs. Every restart re-runs the wait, so a mediocre result plus one retry quietly turns a 5-minute feature into a 15-minute one. Refinement wasn’t in scope before crit. It’s on the list now, because the wait changed the math on every downstream decision.
AI gave product design its waiting room back. The teams that notice first will win the trust of everyone standing in it.
Here’s the industry-wide lens. We spent a decade optimizing waits away, to the point where a spinner past 2 seconds counted as a defect. AI brought the long wait back: generation, conversion, evaluation all take real time, and the honest products admit it. Which means an old craft is current again: progress that tells the truth, exits that don’t lose work, notifications that respect wherever the customer went, and results worth returning to. If your product just grew a 5-minute middle, someone on your team should own it like a page.
The teachable part
When AI adds a long wait, design it like a surface: honest progress, estimated time, a safe exit, a completion signal, and refinement so a bad result doesn’t cost a second wait.