From 80% to shipped: what a 3-5 day finish of an AI-built app actually looks like

Most AI-built apps are not missing features when they stall. They are missing verification.

Something built with Cursor, Lovable, Bolt, v0 or Replit usually reaches a convincing demo fast: screens, flows, a database call or two. Then it sits at "almost ready" for weeks, because the demo ran on the builder's laptop against a preview environment, and every seam that crosses a file, an environment or a point in time was never checked.

I finish those apps for a living. Not by rewriting them - by taking the code as it is, verifying the seams in the environment it will actually run in, and shipping. Here is what a three-to-five day finish looks like, day by day, so you can judge whether it fits what you are holding.
Day 0 - a scope call that removes work

The first call is not a sales call; it is a subtraction exercise. We pick the one job the app must do for its first paying user and list what will not be built this week. Anything that needs a bank contract, a telecom integration or a hardware driver is named now, because those are weeks, not days, and a studio that promises otherwise is guessing.

Output: one paragraph of "done", one list of "not now", and a fixed price for the finish. No deposit before this point.

Day 1 - the seams, verified in production conditions

This is the day that decides the timeline. I run the same list every time, because generated code fails in the same places:
- Sessions. Reload, tab restore and a cold start of the serverless function. Tokens in module memory and secrets regenerated at boot look identical to working auth until the first cold start.
- Ownership. Does every server read and write check who owns the row, or only that a user is logged in? Tested with two accounts, never one.
- Payments. Is a payment confirmed by a redirect page a user can reach by pressing Back, or by a webhook with a verified signature and an idempotency key? One real card payment and one real refund, in live mode.
- Schema. Is the live database the result of a migration tool, or of someone clicking in a dashboard three weeks ago?
- Environment and domains. Does the app crash at boot when a variable is missing, or at the first third-party API call? Are API and front end on the same registrable domain, with cookies and CORS actually agreeing with each other?
- Visibility. When a user says "it broke", is there an error boundary, a request id and one alert channel a human reads?

Each item gets a pass, a fix, or a note that it is out of scope. That list, not a feature list, is the release plan.

Day 2-3 - build only what the list says

Fixes, not features. A working demo goes up at the end of each day on the domain the customer will use, and the customer clicks it. Revisions are free, which is only sustainable because the scope is small and the list is explicit.
Day 4-5 - deploy, then hand over

Deploy is not a button here. Canonical host, HTTPS on both apex and www, a phone on cellular data, a slow connection, and a store submission if the product is a mobile app - privacy manifest, permission strings, an account-deletion path, and a build that does not depend on a debug keystore.

Handover is a short document: where the code lives, what the environment variables are, how to deploy, how to read the logs, and what will break first.

What this is honest about

Three to five days is enough for a focused web or app MVP whose integrations already exist, because the work is verification and finishing, not architecture. It is not enough for a banking integration, a marketplace with two-sided compliance, or a game whose content is still being designed. If your scope is the second kind, the first call should say so - and then the answer is a longer plan, not a faster promise.
I run this as a one-person studio in Seoul: a Korean corporation, a direct contract and an official invoice, the CEO building it with no agency, PM or sales layer in between. Quotes and the review are free; you pay a deposit when we start and the balance only after you have seen the working result.

Studio (English): https://cpcai-studio.netlify.app/en/
Korean: https://cpcai-studio.netlify.app
Contact: yyhak12345@gmail.com

If you built something with an AI tool and it has been "almost ready" for a while, the useful question is not which feature is missing. It is which seam has never been verified outside your laptop.

Comments

Popular posts from this blog

An Australian app build in AUD: GST, ABN and the three decisions to settle on day one

A five-day build is not the impressive part. The handover is.