My actual toolchain: what building looks like day to day
This update was drafted on a schedule by the AI I build with, from real project notes — part of the vibecoding experiment this blog documents.
People ask what I use like there's a secret stack hiding somewhere. Like if they just knew the tools, the seven projects would fall out of their keyboard too. I get why — I used to ask the same question. But the honest answer is boring, and short, and I think the boring part is actually the point.
So here's what's open when I build.
Claude Code. That's the middle of everything. Every single one of these projects — tab., CreatorLens, Confluence, Maison, Fundability, CommunityOS, App Marketing OS — was built with it. Native app or web app, Python engine or React Native, doesn't matter. It's the through-line. If I had to delete everything but one thing, it'd be that. The model underneath it is Claude Opus, and I don't do anything clever there. I just talk to it like it's the fastest engineer I've ever worked with and I'm the one who knows what we're building.
Then it splits by what I'm shipping.
For the native stuff — tab. and Maison — it's Expo and EAS to build, and App Store Connect to actually get it into a human's pocket. That second half is the part nobody warns you about. EAS turns the code into a real signed app; App Store Connect is where you find out Apple has opinions. tab. taught me that one the hard way, but that's a different post.
For the web stuff — CreatorLens, Fundability, CommunityOS — it's Vercel. I push, it deploys, I stop thinking about it. This blog runs the same way. There's no dramatic infrastructure story here and I'm glad there isn't.
Under those, the specific pieces depend on the project, and I try not to be precious about them. tab. is Expo and React Native with Supabase holding the data and RevenueCat handling the subscriptions, plus Sign in with Apple. The web apps are mostly Next.js and TypeScript. App Marketing OS is FastAPI and SQLite because I wanted it dead simple to run, not a cloud bill waiting to happen. Confluence has a Python engine at its core because the math wanted to live there. I pick the thing that gets out of the way, not the thing that looks good in a stack diagram.
And that's genuinely it. There's no tenth secret app. No plugin that writes the product for me while I sleep.
The thing I keep landing on is that the tools stopped being the hard part a while ago. This whole list would've been a real answer to "how do you build fast" a couple years back. Now it's just the setup — the interesting stuff is all the judgment that sits on top of it. What to build. What "good" feels like. When to stop. The toolchain is short on purpose, because the short toolchain is what freed up the room to care about the rest.
This one's auto-drafted from my project notes on a schedule. If a tool isn't actually part of how I build, it's not in here — I'd rather keep the list honest than make it look impressive.