It works in preview, breaks in production
Environment variables, build settings, a database that only ever existed on the preview server.
Fractional CTO · Vibe coding
Stuck vibe coding? I get AI‑built apps to production
You built something real with Lovable, Bolt, v0, Replit, Cursor, Claude Code, Codex or Gemini. It works in the preview. Putting it in front of paying users without fear is the part the AI can’t finish for you, and it’s the part I do.
Sound familiar?
Auth and the database are a mess
Users can see each other’s data, tables were changed by hand, and there are no migrations or backups.
Every fix breaks something else
The AI’s tests pass and the app doesn’t. Each new prompt undoes the last one.
You’re afraid to launch
Keys in the code, a database open to the internet, and no idea what hosting will cost at a thousand users.
What a rescue looks like
Intro call
Thirty minutes, free. Show me the app and where it hurts, and I’ll tell you honestly whether I can help.
Unstick session
Half a day in your code, hosting and data. We fix the worst blocker together, and you get a written list of what comes next, in order.
To production
If you want to keep going: tests that check what your users do, CI/CD with staging, secrets out of the code, and a hosting cost you know in advance.
Healthy software, not a rewrite
I keep what works. The goal is an app that you, your AI tools or your first engineer can change without fear: clear separation of concerns, tests, code review, and deploys so routine they’re boring.
Everything lives in your repository and your accounts. You keep it all when I leave.
Price
Unstick session: half a day, 800 USD. The intro call before it is free.
Bigger jobs, such as a deep dive or a production sprint, are on the fractional CTO rates. Hosting, tokens and tools run on your own accounts.
Questions
Why not spend more on tokens, or hand it to another AI agent?
Try that first; most people I talk to already have. More prompts rarely fix what’s missing, because the gap isn’t code the AI can’t write. It’s decisions nobody made: where the app runs, how it gets deployed, what is tested, and who may see which data. I make those decisions with you, and then your AI tools become useful again.
Do I need to be technical?
No. You need to know your users and their problem. I explain each choice in plain words and write it down.
Will you rewrite my app?
Rarely. Most of what you built is worth keeping; I fix the structure around it.
Is my app safe to launch?
The usual holes in AI-built apps are API keys shipped to the browser, database rules that let any user read every row, no rate limits and no backups. The unstick session checks those first.
Which tools and stacks?
Any AI builder is fine. Most AI-built apps are TypeScript or Python web apps, which is what I work in every day. For mobile: Swift on iOS and Kotlin on Android.
How much will hosting cost?
You get an estimate for your app before we pick a provider, and spending caps on anything that bills by use.
Do you take equity?
Case by case, for part of the fee.
Get unstuck
Tell me what you built and where it’s stuck. The first call is free.
Or send me a message: