How I Ship: Lovable, Claude, and Taste

Here's my stack. Stolen wholesale if it helps you.

Ideation: voice note on my phone while walking my dog. That's it. If the idea isn't clear enough to explain out loud to a 65-pound mutt, it's not clear enough to build.

First build: Lovable. I describe what I want in plain English. Pages, components, what it should feel like. I let Lovable build the whole scaffold in one go. Usually 80% of the way there in 20 minutes.

Refinement: Claude Code. Once Lovable has the skeleton, I clone the repo locally and use Claude Code to do the surgery. Complex state, custom animations, things that need taste more than generation. Claude is my pair programmer. I don't write code alone anymore.

Hosting: Vercel. One-click, auto-deploy on push, never breaks.

Database: Supabase when I need one. Half my apps don't.

Domain: Vercel, Squarespace, or Ionos, whichever one I bought the domain through. I have too many domains. I regret nothing.

Content management: Notion. Everything lives there. Every project has a page. Every brand has a Skills Library.

That's the whole stack. It costs me about $80/month total.

The workflow

Here's what a weekend app looks like.

Friday night: I talk to myself into a voice memo for five to ten minutes. Who is it for? What's the one thing it does? What does it feel like? I don't let myself open a computer until I have answers.

Saturday morning: Lovable. I paste the voice memo transcript as the prompt. I describe the aesthetic (dark, editorial, minimal — I say the same three words every time). Lovable builds the first pass. I live with it for fifteen minutes. Then I start editing.

Saturday afternoon: Content pass. Real copy, real images, real words. Most builds die in the stock-photo-and-lorem-ipsum phase. I don't let mine get there. If I don't have the real thing, I write it.

Sunday morning: Claude Code. Hook up Supabase, add the one clever feature, fix the three things that are bugging me. Ship.

Sunday evening: Tell five people. If three of them use it, it's a thing. If they don't, it's a learning.

What most people get wrong

They wait for the perfect idea. There is no perfect idea. There are ideas you ship and ideas you don't. The shipped ones teach you. The unshipped ones don't.

They over-scope. Every feature you add before launch is a feature that slows you down to launch. Ship the minimum thing. Add features after real people touch it.

They copy-paste AI output without editing. AI gives you the shape. You provide the voice. If you don't edit, your app sounds like every other AI-built app, which is a bad thing to sound like.

They skip the voice memo. This is the biggest one. If you don't know what you're building before you open the tool, the tool will build something confident and wrong, and you'll spend three days trying to fix it.

Clarity up front is the whole unlock.

The part no one tells you

The hardest part of this workflow isn't technical.

It's having the taste to know when something's almost right but not quite, and the nerve to keep editing instead of shipping something mediocre.

Taste is a slow build. It comes from looking at a lot of things and forming opinions. Read sites. Read books. Look at packaging in grocery stores. Notice what makes you feel something and what makes you scroll past.

If you want to ship better, train your eye.

Everything else is just typing.