A Question Picker: The Two States After the First Screen
The first screen did not show the draw
I asked AI to build a question picker for an impromptu speaking session. Questions from four categories would cycle quickly, slow down, and stop on one selection.
The initial design showed only the starting view. The app’s defining moments came during the draw and after it stopped, so I asked to see those states too. Approving the screen before the click would have left the final question—the part a speaker needed to read—unreviewed.
| State | Review question |
|---|---|
| Before the draw | Are the categories and starting control clear? |
| During the draw | Is the question cycling and slowdown visible? |
| After selection | Is the final question easy to read? |
This table summarizes the requests; it is not a set of screenshots from the app.
The event theme did not replace visual preferences
Design feedback continued separately from the interaction review. I asked AI to reflect the event theme while keeping the visual preferences I had already supplied. I disliked the four-category palette and requested Pinterest color references, along with more variation in the circle and semicircle motifs.
For alternatives to the speech-bubble shape, I asked for written descriptions before generating more images. I wanted to compare directions at that stage. No cost or token savings were measured.
Breaking the feedback into palette, repeated motifs, and shape gave AI more specific revision targets than a general request to improve the design.
Animation and selection needed different checks
A smooth slowdown would not prevent a previously used question from appearing again. Reloading the page could also lose the record of questions already selected.
AI reported checking that all 20 English questions could be drawn without repetition and that saved progress could be restored. It then checked desktop and mobile layout, clipping, and animation speed. The final implementation report also included full-screen and reset controls.
These checks addressed different failures. The visual review covered what a speaker would see; exhausting the question list checked whether the rules held after repeated draws.
What reached deployment
The work record includes a deployment report. It does not establish how participants reacted at the event or whether problems occurred during live use.
AI produced designs and code, researched colors, and checked selection rules and layout. I requested the two missing states and corrected the balance between the event theme and my visual preferences. Expanding the review beyond the first screen gave the later feedback a clearer basis.
This account uses records from September 14–15, 2026. The selection, restoration, layout, and deployment checks were reported at the time. I did not rerun the app or investigate the event outcome while writing it.