๐——๐—ฒ๐—ฝ๐—น๐—ผ๐˜†๐—ถ๐—ป๐—ด ๐—”๐—œ-๐—•๐˜‚๐—ถ๐—น๐˜ ๐—”๐—ฝ๐—ฝ๐˜€

You build an app with AI. It works on your computer. You try to put it online. It breaks.

I built VibeSplit. It is an expense splitter with AI scanning. I used Next.js, Prisma, Postgres, and Supabase.

AI makes coding fast. It does not make deployment easy. You need real databases. You need secrets. You need webhooks.

I tested three platforms.

Railway:

Vercel:

Jetpacked:

Watch out for build-time variables. Next.js bakes some values into the browser code. Missing values during the build break your app.

Auth is another trap. Supabase needs production URLs. No platform does this for you. You must fix this in the dashboard.

AI apps are messy. They use many services.

You need tools to understand the whole shape of your app. Framework names are not enough.

This is the bar for AI tools. They must understand the full service stack.

Source: https://dev.to/thpl/deploying-a-complex-cursor-built-app-prisma-postgres-inngest-supabase-in-production-4019