Vibe-coded apps are piling up. Here is how to keep them maintainable.

Vibe-coded apps are piling up. Here is how to keep them maintainable.Building software has never been easier. Keeping it has never been more expensive to ignore.

Want this running in your business?

Maxpertise builds AI workflows you own, live in about ten days.

Send a written intake · What we do

The "vibe coding" wave - Lovable, v0, Replit, Bolt - let non-engineers generate working apps from prompts. Bubble, the no-code platform, leaned into the aftermath this month: its October 26 AMA demonstrated importing apps built with those tools (including zip-based imports) so they can be edited visually, and its October 6 Agent 2 launch added an MCP connection letting Claude Code and Codex act on Bubble apps. When a major platform builds an import pipeline for other tools' output, that tells you the market has moved from "can I build it?" to "who maintains it?"That second question is about to hit thousands of small businesses at once. This post is the checklist for being ready before it does.What actually goes wrong with vibe-coded appsAI-generated apps work - at first. The failure modes arrive on a predictable schedule:

Week 2-4: the change request. "Can it also send an SMS?" The prompt that built v1 does not safely modify v1. Regenerating from scratch loses the tweaks you made since; hand-editing generated code without understanding it breaks things you didn't know existed.

Month 2: the data problem. The app stores customer data in a structure nobody reviewed. Now you need reporting, or a migration, or GDPR deletion, and the schema was never designed for any of it.

Month 3-6: the integration debt. The app talks to nothing - not your accounting, not your CRM. It was a demo that became operations, and every manual re-entry between it and your real systems is a slow leak.

Any time: the owner-dependency trap. The one person who knew the prompts and the workarounds moves on. The app outlives them and nothing else.

The pattern is identical to what the software industry has dealt with forever - legacy code nobody understands - except vibe-coded apps reach that state in months instead of years, and the businesses holding them rarely have developers on staff.The maintenance checklist for an AI-built appIf you have a vibe-coded app in production (or about to be), run this list. It takes an hour and tells you whether the app is an asset or a liability:

Can you rebuild it? Export the code (zip, GitHub - whatever the tool offers). If the app exists only inside the vendor's editor, you do not own it; one price change or shutdown and it is gone. Ownership of code and data is non-negotiable for anything business-critical - it is the first thing we check in any build, ours included.

Who reviews changes? Every change to the app needs a second pair of eyes, even if the reviewer is external. AI generation plus no review is how a silent data bug becomes a customer-facing incident.

Where does the data live, and is it backed up? Verify the database is exportable and actually backed up on a schedule. Test a restore once. An untested backup is a hope, not a backup.

What does it cost when it succeeds? Usage-priced AI features inside the app can spike - the same mechanism behind SMB AI bill shock. Put a cap on anything metered.

Is there a written "how this works" page? One page: what the app does, which services it touches, where credentials live, who to call when it breaks. This is the difference between an app and a mystery.

If the app fails two or more of these, it is accumulating debt faster than value.Own the workflow, not just the codeThe deeper fix is a mindset shift. The vibe-coding tools sell the artifact - an app. What a business actually needs is the workflow: leads flow in, data lands in the right systems, someone reviews what matters, nothing depends on a single prompt.That is why the platform moves this month matter. Bubble's import pipeline and MCP connection acknowledge that apps will be rebuilt, migrated and maintained across tools - the app is not the asset, the running process is. The same logic we apply to agents and automation for small business: the deliverable is a process that keeps working when the hype moves on.Practically: when you outgrow a vibe-coded app, the sequence that works is (1) export everything - code and data, (2) map what the app actually does against your business process, (3) rebuild the workflow properly with review points and integrations designed in, (4) migrate the data with a parallel run, (5) keep the old app frozen as reference until the new one proves itself. That is a two-to-three-week project when done deliberately - and an emergency when done reactively after a failure.When vibe coding is the right tool anywayTo be fair to the wave: for internal tools, prototypes, and short-lived experiments, vibe coding is excellent. If the app's job is "validate this idea for two weeks," the economics favor building fast and discarding. The trouble starts when a prototype quietly becomes production - when real customers are in the data and the business depends on it. That transition, prototype to production, is the exact moment to apply the checklist above or rebuild properly.Frequently asked questionsCan I just keep prompting the AI tool to maintain the app?For small changes, sometimes. But each regeneration risks breaking what worked, and the tool has no memory of your business rules. Past a certain size, prompt-driven maintenance costs more than a proper rebuild.Is Bubble's import feature worth using?It is a genuine option for apps with modest logic - visual editing beats raw generated code for non-engineers. Check what the import does to your data model and integrations before committing; imports rarely carry those over cleanly.How do I know if my app is "production"?Simple test: would its failure cost you money or customers this week? If yes, it is production, and it needs review, backup, ownership, and a maintenance plan.What does a proper rebuild cost?Usually less than expected, because the vibe-coded version is a working spec - it shows exactly what the workflow needs. Fixed-price rebuilds of a scoped app typically run days to a few weeks, not months.The bottom lineThe vibe-coding wave solved the wrong half of the problem. Building is now nearly free; maintaining is where the money is - and where the platforms are now racing to sell you a solution. Small businesses that treat AI-built apps as disposable prototypes are fine. The ones that let a prototype become critical infrastructure without ownership, review and backup are writing next year's emergency.If you have an AI-built app that has quietly become important, send us a written intake - we will run the maintenance checklist with you and tell you honestly whether to keep, migrate, or rebuild.

Maxpertise is an AI-native engineering company. We embed native AI engineers inside your team, live in about 10 days. Please enable JavaScript to view the site, or email contact@maxpertise.net.