I was staring at 500 wallpapers and hitting the same wall I always hit.
I've been building iOS wallpaper apps on my own since 2014. Across those 12 years, the apps have accumulated over 50 million downloads — a number I'm still quietly proud of. But every time I add a new collection, I face the same problem: how do I categorize these images?
This time, with 500+ new images ready to ship, I'd spent two hours going in circles. So I did something different. I asked Claude.
I described the images in text, shared my existing category structure, and asked: "What categories would help users find images most intuitively?"
My "Correct" Structure vs. What Users Actually Feel
What came back surprised me.
My categories had always been organized by color (Blue Tones, Earth Colors) and subject matter (Nature, Urban, Abstract). This felt logical — I had informal data built up over years of watching download patterns suggesting users browse by visual attribute.
Claude proposed something different. Categories organized by mood ("When you want to feel calm," "When you need focus") and context ("Good for mornings," "For winding down before sleep").
My first reaction was skeptical. This feels too soft, too vague. But sitting with it for a few minutes, I realized something I'd missed for ten years:
People don't search for "a blue image." They search for something that matches how they feel right now.
I'd spent a decade organizing images by their visual properties. But users might be browsing from the inside out — by mood, by the moment in their day, not by what the image looks like.
The Gap Between Creator Logic and User Intuition
This made me think about my grandfathers.
Both were miyadaiku — traditional temple carpenters who spent their lives building and restoring Shinto shrines and Buddhist temples. They would often say that the work wasn't really about technical precision. It was about understanding how a person would move through the space, how light would shift with the seasons, what someone would feel standing in the entrance at dawn.
I always thought I was doing something similar with my apps — designing from the user's experience outward. But this conversation with Claude revealed a gap I hadn't noticed. My color-based taxonomy was built from the inside out, from my perspective as the person who made the images. The mood-based approach came from the outside in, from how users actually encounter their phone screen throughout the day.
I was giving users my map. They might have needed a different one.
What Actually Changed
I tested it. For the new collection, I kept the familiar color and subject categories while layering in scene-based ones: Morning Light, Before Sleep, Focus Mode, Weekend Calm.
After a few weeks, the reviews shifted in a subtle but noticeable way. Before: "Can you add more blue wallpapers?" After: "The Before Sleep collection is perfect." "I open Morning Light every single morning now."
Whether this is correlation or causation, I can't say for certain. But users started describing their relationship to the wallpapers differently — not as visual objects they'd selected, but as companions to specific parts of their day.
Claude Didn't Give Me "The Answer"
I want to be clear: Claude's suggestion wasn't some objectively correct framework that I should have discovered years ago. Color-based and subject-based categorization still works. The scene-based approach isn't universally better.
What Claude gave me was a second map to look at alongside my own. It didn't dismiss the decade of pattern recognition I'd built up. It just quietly offered another angle. Deciding what to do with that was still my call.
This is the specific value I've come to appreciate. As a solo developer, the assumptions you carry never get challenged before release. Reviews tell you what's wrong after the decision is already shipped. In the design phase, the chances of someone asking "but is this really how users think about it?" are close to zero.
Claude can be that voice — not as a replacement for experience, but as a conversation partner that brings a different perspective without needing you to explain yourself defensively first.
On "Breaking" a 10-Year Habit
The phrase I used in the title — "shattered an assumption" — isn't regret. It's actually the opposite.
You can only break something if you built it first. The ten years of data, of watching what users download, of iterating based on reviews — all of that is what made this conversation useful. Without it, Claude's suggestion would have been just an interesting idea. With it, I had the foundation to actually evaluate whether a different framing made sense.
My grandfathers would refine a technique over years, then deliberately set it aside to look at the work with fresh eyes. That tension between accumulated skill and beginner's curiosity was, I think, how they kept building things that felt alive rather than just technically correct.
I'm trying to bring something like that into how I use Claude — not as a shortcut past experience, but as a way to keep questioning the frames I've built up so quietly that I stopped noticing them.
If you've had a similar moment where a conversation with an AI (or anyone) helped you see a blind spot in your own work, I'd genuinely like to hear about it.