Memoria

A clearer way to finish recording

In testing, one beta user in their 60s pressed Pause when they meant to finish a recording. However, they got stuck between Pause and Resume, unsure how to end the recording.

The original flow had been built with AI around a common recording pattern, but what I observed indicated the paused state didn't make the exit path clear enough.

Reasonable at first, but unclear after pause.

Based on this observation, I realized the paused state needed a clearer way to finish the recording. I already had a direction in mind — surface a "Done" action near the main controls — and used Claude Code to quickly generate a first-pass fix without over-specifying the details.

The first AI pass solved the immediate issue, but the interaction felt rough.

Functionally, it worked. But the interaction still felt unpolished — the buttons shifted too much, and the oversized Done action made the paused state feel visually uneven.

I then used what Claude Code generated as a base, and sketched in Figma to guide the next move. I explored two directions: one removed the top-right "Done" entirely and always showed two buttons, making the next step constantly visible; the other kept the top-right action as a consistent anchor, while adding a contextual "Done" after pause.

Direction 1 — removed the top-right "Done" and kept a persistent "Done" button near Pause/Resume.

Direction 2 — kept top-right "Done" and added a second "Done" action after pause.

I chose direction 2 because it created a better balance between clarity, familiarity, and efficiency:

• Kept a familiar anchor: the top-right "Done" remained available for users who already knew that pattern.
• Made the next step clearer: a second "Done" appeared after pause, exactly when users were most likely to need it.
• Reduced visual overload: unlike direction 1, it did not show two actions all the time.
• Avoided extra friction: unlike direction 1, it did not turn a one-step action into a forced two-step flow.

The next AI pass got the structure mostly right, but the visual execution still needed work.

AI output — structure improved, but visual execution was still rough.

Based on this implementation, I created a calibration sketch to correct the alignment and improve the overall visual balance.

With the main issue largely resolved, a subtler problem became easier to notice: "Done" felt too vague in this context. I changed it to "Finish" to make the completion signal clearer and more explicit.

My refinements — alignment, icon weight, spacing, and label clarity.

After these refinements, ending a recording became a much clearer next step. In later testing, I no longer observed the same confusion in the recording flow.

AI got to a workable first pass quickly. What still needed a designer was judgment — what to keep, what to change, what to rename, and what still didn't feel right. That judgment is what turned a functional fix into a smoother recording flow.