95%
Barbara·5 min read

About 95% of Replit apps never reach production. Where AI-built apps get stuck

The five places AI-built apps stall after the preview, and how we designed Leanfinit around them.

0Mapps built on Replit
0Kwent live
0/54Apple developers has ever shipped an app

About 95% of Replit apps never reached production: roughly 5 million were built, and about 250,000 went live. The generation step is no longer what stops people. Apps stall after the preview, at hosting, data, accounts, stores and upkeep, the work that has nothing to do with the idea.

That number shaped how we built Leanfinit. Below are the five places where we see apps stop, and what each one asks of the person who only wanted a working tool.

01

The preview is not the product

A preview proves that the app renders. It does not prove that the app will still work tomorrow, on someone else's phone, with real data in it.

Most AI builders are tuned to make that first preview impressive, because that is the moment someone decides to keep going. The trouble starts on the next screen, where "it works" turns into a list of decisions that a developer makes without noticing and a non-developer has never had to make.

02

Stuck point 1: where the app lives

A web app needs hosting, a domain and a deployment step. For a developer that is routine. For someone who has never heard the word "deploy", it is a wall made of unfamiliar dashboards, DNS records and billing pages.

Every extra account between the preview and a link you can send is a place where people stop. The fix is not a better tutorial. It is fewer steps.

That is why the Publish button in our studio does one thing: it puts the app live on the web, so there is a link to open and share. You can read what happens when you press Publish in detail.

03

Stuck point 2: where the data goes

The second wall is data. Most generated apps assume a database on a server: a provider to sign up for, keys to paste, tables to create, rules to set so strangers can't read your entries.

This is where a lot of personal apps die. A habit tracker or a budget for one person does not need a server. It needs a place to keep entries that is private and still there next week.

We made the opposite default. Apps built with Leanfinit keep their data on the device, with no backend by default. There is nothing to provision, nothing to secure on a server and no monthly database bill. The trade-off is real: data on one device is not shared with other people. For many personal tools, that is the right trade.

04

Stuck point 3: accounts, logins and keys

AI-built apps often arrive with a sign-up screen nobody asked for, because "users and accounts" is the default pattern in the code the model learned from. Then come the auth provider, the email templates and the password reset flow.

For an app used by one person or one small team, every login screen is friction with no benefit. And every API key pasted into a settings page is a security question the person can't answer.

The rule we follow: an app should not ask for an account unless the job needs one.

05

Stuck point 4: the app stores

For mobile apps, the store is its own project. Apple's developer program, review guidelines, screenshots, privacy labels, test builds, and a rejection email in language written for engineers. Fewer than 1 in 54 registered Apple developers has shipped an app, and those are people who already chose to become developers.

We are honest about where we stand: during the beta, Leanfinit does not submit apps to the App Store or Google Play. You publish to the web. If you need a store listing, you can export the code and take that route yourself. We wrote about why so few developers ship to the App Store and where a personal app should live.

Many people discover they never needed a store at all. If the app is for you, your family or your shop, a link often does the job. Our insight on making a simple app for yourself without publishing to the App Store covers that case.

06

Stuck point 5: keeping it alive

The last wall arrives months later. A library updates, a provider changes its free tier, a key expires, and the app breaks. The person who prompted it can't read the code, so the app quietly stops being used.

Fewer moving parts means fewer things that break on their own. An app with no server, no third-party database and no login service has very little that can expire under it. When you want to change something, you describe the change instead of debugging someone else's code.

07

What we took from it

Generation got cheap. Everything around it stayed expensive. So we judge our own product less by how good the first preview looks and more by whether the app is still useful a month later.

That is the reasoning behind the choices above: free during the beta, so cost is not the reason someone stops; data on the device, so there is no database to set up; one Publish button, so there is a link at the end; code export, so nobody is locked in when they outgrow us.

If you want to compare tools by this standard, our guide to free app builders without coding looks at which ones actually let you get to a live app.

Try it on a small problem of your own in the Leanfinit studio.