Case study · design task · 2026

File search for a billion-file browser: before and after feedback

I designed a search form, got it rejected for being too complex, and rebuilt it as one bar of conditions above the file list.

Before what I sent

A side panel with three numbered steps, a count under every condition, an AI field, and the whole file browser around it.

Search panel with three steps, AI field, two conditions with counts, and a full file browser around it
The form competes with the tree, the panel header, the AI field and the counts.
Search results with a notice, list and grid toggle, and the panel still open
Results, a notice and the panel at once: the search is no longer the focus.

Feedback from the reviewers, and what it changed

Reviewers saidChange
Side elements drew more attention than the searchOne bar above the list. No panel, steps, counts or results.
A custom extension was very hard to searchType “.cram” and press Enter: “Type is .cram”. Any extension works.
The AI assistant felt bolted onRemoved. The brief asked to build with AI, not to add AI.
Location was shown twice and changed in the less visible placeOne chip, first in the bar. It shows the location and changes it.
Research was too long; the prototype was hard to reachThis page, and one link to each prototype.

After the core task in the fewest steps

One search bar with chips: In 2023-wgs, Type .pdf, Size over 100 MB, Not modified for 1 year, and a Search button
The brief’s example: three chips and Search. The chips read as one sentence.
Typing .cram suggests Type is .cram
Type any extension; Enter adds it.
Small editor for the Type condition with extension suggestions
Click a chip to edit it; × removes it.
Searching all storage with a Cancel button
Search starts a searching state with Cancel. The flow ends here, as the brief asks.

What I learned

  1. Restraint shows maturity. Every extra weakened the proof that the user reaches the goal fast.
  2. Limiting words are a ceiling. “Simple” and “a few basic conditions” should have cut the scope on day one.
  3. Solve the brief’s own example first, then its nearest variant: any extension, not only PDF.
  4. One fact, one place, editable where it is shown.
  5. With AI tools, my job is editing. AI made adding cheap; I set a budget and cut.
  6. Five minutes with one fresh person would have caught the extension and location problems.

Client details are anonymized; file names and data are invented. Built on my design system Big Hat; prototypes built with Claude Code.