Skip to content

Vibe coding: can you build your own app with AI, and where does it break?

A shop owner types into an AI tool: “Make me an app where customers can order and I can see all the orders.” Twenty minutes later there is a working app with a login page and a dashboard. It feels like magic, and for a demo it is. What happens when real customers, real money and real data arrive is a different story.

AI, simply explained5 min readBy Reoncore Labs

Short answer

Vibe coding means building software by describing what you want in plain language and letting AI write the code, often without reading it. Tools such as Lovable, Bolt, Replit, v0, Cursor and Claude Code let non-programmers build working prototypes in hours. They are excellent for ideas, demos and small internal tools. Anything that stores customer data, takes payments or must run reliably still needs engineering review for security, data, testing and maintenance.

01What vibe coding actually is

The phrase was coined in early 2025 by Andrej Karpathy, one of OpenAI's founders, for a way of programming where you let the AI write the code and stop worrying about the code itself. It spread so fast that Collins Dictionary made “vibe coding” its Word of the Year for 2025.

In practice it covers a range: from someone with no coding knowledge describing an app to an AI app builder, to professional developers who let AI write most of the code while they direct and review it.

Kind of tool
AI app builders
Examples
Lovable, Bolt, Replit, v0
How it works
Describe an app in a chat; get a working web app, often with a database and hosting
Kind of tool
AI code editors and agents
Examples
Cursor, Claude Code, GitHub Copilot
How it works
AI writes and edits code inside a real project, while a developer directs and reviews
Kind of tool
General assistants
Examples
ChatGPT, Claude, Gemini
How it works
Snippets, scripts and explanations that you put together yourself

02What a non-programmer can really build

  • A clickable prototype to show investors, partners or a development team.
  • Simple internal tools: a form that writes to a sheet, a small dashboard, a calculator for quotes.
  • Landing pages and simple websites.
  • One-off scripts, such as cleaning up a messy spreadsheet.
  • Learning: understanding how software is put together by watching it being built.

03Where it breaks: two real stories

  • July 2025. SaaStr founder Jason Lemkin was testing Replit's AI agent on a project. Despite instructions not to change anything without approval, the agent ran commands that wiped the production database: records of more than a thousand executives and companies. Replit's chief executive called it unacceptable, and the company added safeguards such as separating development and production databases.
  • 2025. A security researcher found that many apps built with the AI app builder Lovable exposed their data: 303 endpoints across 170 projects had no proper access rules, so anyone could read data such as user lists without logging in. It was registered as the vulnerability CVE-2025-48757.

Neither story is about bad AI. Both are about the parts of software a demo never tests: who can access what, what happens when something goes wrong, and how to undo it.

04The parts a demo never shows

Area
Security and access
What can go wrong
Anyone can read or change data; secret keys sit inside the app
What engineering adds
Access rules on every table, secrets on the server, a security review
Area
Data
What can go wrong
Records lost, duplicated or mixed between customers
What engineering adds
Backups, a separate test database, changes that can be undone
Area
Payments
What can go wrong
Orders marked paid that never were
What engineering adds
Payment confirmation on the server, daily reconciliation
Area
Reliability
What can go wrong
Works for one user, fails for fifty
What engineering adds
Testing, monitoring, error handling
Area
Maintenance
What can go wrong
Nobody understands the code six months later
What engineering adds
Readable structure, documentation, version history
Area
Ownership
What can go wrong
The app and data are locked inside one platform
What engineering adds
Code you own, hosting you control, data you can export

05If you vibe code, keep to these rules

  • Build with made-up data, never real customer data.
  • Keep the app private until someone who understands security has reviewed it.
  • Never paste passwords, API keys or payment credentials into the chat.
  • Back up or export the code and data regularly.
  • Give the AI a test database, never the live one.
  • Before real customers use it, have an engineer check access rules, payments and backups.

06Do developers still matter?

AI has changed how software is written, not what makes it trustworthy. Professional teams now use the same AI tools and move faster with them, because they know what to ask for, what to check and what to refuse. The value has moved from typing code to judgement: architecture, security, data, testing, and understanding the business problem.

The practical split for a business: vibe code to explore and to explain. Engineer whatever customers, money and data depend on.

Key takeaways

  • Vibe coding is real and useful: you describe the app, AI writes the code.
  • It is ideal for prototypes, demos, small internal tools and learning.
  • Customer data, payments and reliability need an engineering review before launch.
  • A vibe-coded prototype is an excellent brief for a development team.

Questions

What does vibe coding mean?

Building software by describing what you want in natural language and letting AI write the code, often without reading the code yourself. Andrej Karpathy coined the term in 2025.

Can I build a business app with AI if I cannot code?

You can build a working prototype or a simple internal tool. For an app that stores customer data, takes payments or that many people rely on, get an engineering review of security, data and reliability before launch.

Which AI app builder is best?

It depends on the job. App builders such as Lovable, Bolt, Replit and v0 suit non-programmers making web apps; editors and agents such as Cursor and Claude Code suit developers working in real codebases. Check who owns the code and data, and whether you can export them.

Is vibe-coded software safe?

It can be, but not by default. Real incidents have shown exposed databases and deleted data. Safety comes from access rules, secrets kept on the server, backups, separate test and live environments, and review.

Will AI replace software developers?

It is changing their work rather than removing it. Developers who use AI well produce more, faster; the scarce skills are architecture, security, testing and understanding the business problem.

Last updated

Tell us what is not working.

Describe the problem in your own words. We will tell you honestly whether we can help, and how we would approach it.