When the AI-built app still needs a real developer

I keep seeing the same nearly-finished app. Someone prompted an AI, got screens and a repo that kind of runs, then hit the wall. Logins half work. Payments are "coming soon". Deploy is a mystery. The founder knows enough to demo it and not enough to ship it safely.

That is not a failure of curiosity. It is what happens when the first draft looks like a product and the last 20% is the real product. I call that a vibe-code rescue. You still want the idea. You need someone who builds for production to finish it.

Where AI-built apps usually stall

The demos look fine. Click paths work on the happy path. Then a real user signs up twice, a webhook fails quietly, or staging and live share the same Stripe keys. Or there is no staging at all.

Common sticky spots I walk into:

• Auth that works until password reset, OAuth, or roles show up.
• Payments that charge once in test mode and fall over on live keys, webhooks, or refunds.
• Database migrations nobody wants to run twice, or seed data treated as production truth.
• No backups, no error tracking, no queue for emails, and a single server that was never meant to stay up.
• Prompt-generated code that works until the next change, because nobody owns the structure.

None of that means start again from zero. It means treat the repo as a draft and decide what must be solid before customers rely on it.

What a rescue actually looks like

I do not polish every line the AI wrote for fun. I find the path that has to work for your business, then make that path boring and reliable.

Usually that means a short audit of what exists, a clear list of blockers, then focused build work. Fix auth and billing. Put staging somewhere real. Wire webhooks. Add logging so failures leave a trail. Tighten the bits that hold money, data, or logins. Leave the rest alone until it matters.

Sometimes the right call is keep the idea and rewrite a thin slice. Sometimes it is finish what is there. The point is a product you can run, not another demo that dies when the prompt window closes.

When this beats another rebuild quote

A full rebuild is the expensive answer when nobody has looked at what already works. A technical review helps when the problem is unclear. A vibe-code rescue is for when the problem is clear: AI got you most of the way, and you need a developer to own the finish.

That suits founders who built with Cursor, ChatGPT, Lovable-style tools, or a mix of no-code and generated code. It also suits agencies whose client arrived with a half-working AI repo and a launch date.

I price a vibe-code rescue from £495 for a defined rescue pass. Bigger apps become scoped project work. Details sit on the service page. Have a look if the description matches what you are staring at.

What this is not

It is not a free tutorial on prompting better. Prompting will not fix production auth, live payments, or a deploy with no rollback. It is also not a promise that every AI line stays. Some of it will. Some of it will not. Honesty beats protecting a repo that was never meant to scale.

If the app is further along and you need ongoing ownership rather than a one-off finish, I also do Laravel systems and ongoing support.

Vibe-code rescue from £495 →

Almost working is a painful place to sit. Getting the last stretch into something you can trust is usually cheaper than pretending another prompt will finish it. Cheers.