Solo-building a real LA dentist directory in public. 598 dentists, on Claude Code + Supabase. The wins, costs, and flops, daily. Day one, near-zero traffic.
💸 I’m paying for two AI builders this month… and I haven’t touched either.
Bolt: 10.6M tokens unused
Emergent: 110 credits sitting there
I’d rather ship one small, genuinely useful thing than watch the credits disappear.
What should I build next? Drop your best idea below and I’ll pick one and build it in public, sharing every step.
#BuildInPublic
Working as a promotional products, signage and custom apparel company. Our suppliers are catching up with the times and they started with the claude MCP. Lucky me!
@gohighlevel Do you feel it's better to pick one tool and excel in it when you're just beginning or try to experiment with multiple tools to find one that works well for your businesses? I'm not sure exactly where to start.
Shopify gotcha that cost me an hour: creating a collection does not publish it to your headless sales channel. The API call succeeds, then your storefront only sees the default one. You have to run publishablePublish for each collection. Products will need the same step.
Brand rule I set for my apparel label: no em dashes. Anywhere.
It's the loudest tell that a machine wrote your copy. Swept 16 blog posts, 6 page files, 33 pin descriptions and every alt text down to zero.
AI writes plenty for that brand. It doesn't get to sound like one.
@LiuweijiaVip Its so much cheaper and easier to brainstorm with claude. Then I use bolt and emergent to build the framework. Emergent is great with UI stuff.
Cheapest way to use an AI app builder: stop thinking inside it.
The spec, the data model, the edge cases all got argued out in Claude Code, where iterating costs pennies. Emergent got one clean prompt and built the UI. Wiring it into the real app came home to my repo.
Set up the database for my review system at 2am. Five tables, row level security on every one as a deny-all second wall. The server already talks through its own key, so if that key ever leaks the tables still say no. The migration lives in the repo like any other code.
Mobile LCP was 9.4s. Desktop was fine.
My useIsMobile hook starts false, so the server shipped desktop markup. Phones threw it away at hydration and Chrome stamped LCP on the freshly recreated h1.
Fix: render both layouts, gate with CSS, prune after mount. LCP now equals FCP.
Found a race in a review list: change the sort while page 2 is loading and the old sort's results get appended to the new list. Fix is a request sequence counter. Every response checks if it is still the latest before touching state. Stale responses get dropped.
My review system is live in production right now and nobody can see it. Feature flag off means the routes 404 and the homepage is byte identical to before. 278 tests green, flag on verified locally. Shipping dark beats a long-lived branch that rots.
Built an editorial lookbook for my storefront without burning a month of AI credits.
Spec written locally with Claude first. Emergent got one prompt, front end only, against a mock of my real data types. I ported it into the repo myself.
Vague prompts are what cost money.
@Aditya_181105 Always frontend unless building a very specific piece of work. You never know what front end doorways you want to add in if you go backend first then you have to rewrite the backend anyway