When your Bolt.new project outgrows Bolt
Bolt.new is a wonderful way to figure out what you want to build. You describe an app, and it builds and runs the whole thing right in your browser. For the first few weeks, it feels like magic.
Then the project gets bigger, and the magic starts to slow down. Here’s how to tell when you’ve hit that point, and what to do about it.
The signs
- Every change costs more. Each prompt uses more tokens than the last, because there’s more code for the AI to read.
- The AI forgets earlier decisions. It reintroduces bugs you fixed last week, or rewrites something that was working.
- The backend is half done. Supabase is connected, but the security rules, relationships between tables and migrations aren’t there.
- It works in Bolt and nowhere else. Moving it to real hosting surfaces missing settings and build errors.
- You’re nervous about every change. There’s no easy way to see what changed or roll it back.
If three or more of those sound familiar, you haven’t failed. You’ve built enough of a product to need a different way of working.
Why it happens
AI builders work best when the whole project fits in the AI’s head at once. Small apps do. Real products don’t. Past a certain size, the AI can only see part of the code at a time, so its changes start to contradict each other.
The fix isn’t a better prompt. It’s structure: a proper code repository, tests that catch regressions, and someone who understands the whole system.
How to move it out
1. Get it into Git
Export the project to GitHub. From then on, every change is recorded, can be reviewed, and can be undone. This step alone takes away a lot of the anxiety.
2. Make it build anywhere
Get the app building and running outside Bolt, on your own machine and on real hosting. This is where missing environment variables and hidden dependencies show up, and it’s better to find them now than on launch day.
3. Finish the backend
If you’re on Supabase, set up row-level security on every table, define the relationships between them, and record the schema as migrations. (Here’s a plain-English guide to row-level security.)
4. Make the integrations handle failure
Payments, email and other APIs need to cope when things go wrong: a declined card, a timeout, a service that’s down. Bolt projects often handle only the happy path.
5. Speed it up
Large bundles, uncompressed images and too many database calls make apps slow on real phones and networks. A short performance pass makes a big difference.
Can you keep using Bolt?
For small changes, you can. But once the app is in a proper repository with tests, most teams find normal development faster and safer, often still with AI tools, just with more structure around them.
You don’t lose anything by moving. The screens, the flows and everything you learned building them all come with you.
If you’re not sure the app is worth moving at all, see rescue or rebuild your AI-built app. For the security gaps to close on the way, see vibe coding security risks.
If your Bolt project has hit the wall, our Bolt rescue takes it from there: we export it, finish the backend and get it ready for real users.
