Cross-posted from Shipped With AI, my build log about learning to ship small tools with AI coding assistants. Originally published August 29, 2026, as the sixth post on the site.
Status check, up front: this is a bug fix to a prototype, not a client story. No shop has signed up to use this tool. What changed is that a limitation I described in an earlier post no longer exists.
In an earlier post about the product catalog tool, I admitted something that wasn't finished: the "fushi" version — built for furniture and craft workshops — splits the owner's editing tool and the public catalog into two separate HTML files. The editing tool had a live preview panel that reflected changes instantly, but that preview only updated the tool's own screen. The separate public catalog page never saw those changes. An owner could edit a product for as long as they wanted and the page a customer actually visits wouldn't move.
I called that "a real limitation, not a design choice" and said it was next on the list. This post is that next step.
What was actually wrong
Both files had their own hardcoded product list sitting directly in the page's JavaScript. The update tool's list and the catalog page's list were two independent copies of the same idea, with no connection between them. Editing one never touched the other — there was nothing to sync in the first place, because nothing was shared.
What I changed
- A shared save location. Both files now read and write the same browser storage key (localStorage). The update tool already saved to it; the catalog page previously ignored it and just used its own hardcoded list. Now the catalog page loads from that same key first, and only falls back to its original hardcoded list as seed data if nothing has been saved yet.
-
A publish flag, so editing isn't the same as going live. Every product now carries a
publishedflag. The public catalog filters out anything not explicitly marked published. That matters because without it, saving a half-written product description would have made it visible to customers the moment it was saved — the fix for "changes don't show up" shouldn't accidentally create "changes show up too early." -
Same-tab-open sync. If a shop owner keeps the public catalog open in one browser tab while editing in another, the catalog tab now listens for the browser's
storageevent and re-renders itself automatically when the update tool saves — no manual refresh needed.
How I checked it actually worked
Code that looks correct and code that behaves correctly aren't always the same thing, especially for something involving two separate files and a browser storage event. Instead of reading the change and assuming it was fine, I ran both pages in a headless browser and checked two things directly: that saving an edit in the update tool actually shows up on the catalog page, and that a product left as a draft actually stays hidden. Both held up. That's a small habit, but it's the same one that caught a mismatched data column in one of my note.com posts before it went out — check the actual behavior, not the intention.
Why this sat unfixed for a while
Nothing forced this fix to happen when it did — there's still no shop using the tool, so nothing was actually broken for a real user. What changed was that I'd already written, in public, that this was a known gap and "next on the list." Having said that in a post that's still live made it a specific thing to go back and close, instead of a vague someday-improvement that could keep sliding.
Where this actually stands
The "fushi" catalog prototype now behaves the way I originally described it as behaving, instead of the way it actually behaved. That's the whole update — no new features, no client, no revenue. Across everything I'm running (this site, the product catalog business, and posting on note.com), the honest total is still zero dollars and zero yen, same as when I wrote about the system deciding what to build next. What's different is that one specific, named gap between "what I said it does" and "what it does" is now closed.
If you're curious about the actual tools this same workflow has produced — a pricing calculator, an image/PDF processor, a product catalog builder — they're on Shipped With AI, along with the rest of this honest, zero-revenue-so-far build log. Written by a student in Japan, edited with the help of AI tools.
Top comments (0)