Imagine you’ve always wanted your own app. Perhaps one that tracks your household budget, organises the family shopping list or reminds you when to water your plants.
Until recently, turning that idea into reality would have meant learning to code or paying a software developer. Now you can simply tell an artificial intelligence (AI) chatbot about your idea. Within minutes, it can write the code, fix errors and help you build a working app.
This approach is known as vibe coding. It’s fast, exciting and is making it possible for millions of people to create software for the first time.
It’s easy to see why vibe coding has become so popular. Yet researchers have already found thousands of security flaws and exposed secrets in publicly available AI-generated applications. Software that appears to work isn’t always software you can trust.
What is vibe coding?
The term “vibe coding” was coined by AI researcher Andrej Karpathy. Rather than writing the software yourself, you describe the outcome you want – or the “vibe” – in plain, everyday language and let AI generate most or all of the software.
If something doesn’t work, you ask the AI tool to fix it. You don’t necessarily need to understand how the code works – you just keep prompting until you get the result you’re after.
Tools such as ChatGPT, Claude, Gemini and specialised generative AI-based coding assistants have made this possible. In fact, the most recent survey by major coding platform Stack Overflow revealed that 84% of developers globally used or planned to use AI coding tools in 2025.
And it’s not just developers using them: recent stats indicate about 63% of users of vibe coding platforms are people without a programming background, building products and tools of their own.
People with little or no programming experience can now build websites, apps and automated workflows (such as compiling documents or doing other complex tasks) in a fraction of the time it once took.
Building reliable software
Most people imagine software developers spend their days writing code. In reality, coding is only one part of a much bigger process.
Before software is released, developers ask important questions. Does it work reliably? Is it secure? Can someone else maintain it? What happens if something goes wrong?
They test different situations, review each other’s work and fix problems before anyone else uses the software.
These steps aren’t there to slow developers down. They’re there because even software that seems to work for its purpose can still expose sensitive data or disrupt critical systems.
Vibe coding can make it tempting to skip much of this process. If the app appears to work, it’s easy to assume the job is finished.
The risks of vibe coding
Now, suppose you ask AI to build an app to manage your household expenses.
The app opens, records your purchases and calculates your spending. Everything seems fine.
But is it calculating correctly in every situation? Is your financial information stored securely? What happens if the app crashes halfway through entering your monthly expenses?
Or perhaps it stores your financial data in a way that’s easy for others to access. The app does exactly what you asked – but your personal information may not be properly protected.
These aren’t questions most beginners think to ask, yet they’re exactly the kinds of questions professional developers routinely consider.
Fixing problems can also become frustrating. You ask the AI to solve one issue, only to discover it has introduced another somewhere else. Without understanding what the AI has changed, you’re left relying on it to solve problems it may have created.
When is vibe coding enough?
None of this means vibe coding should be avoided entirely.
It’s an excellent way to experiment with ideas, learn programming concepts or quickly build personal tools. Entrepreneurs can create prototypes, hobbyists can automate repetitive tasks, and small organisations can explore ideas that might otherwise never get built.
The risks increase when AI-generated software is treated as ready for real-world use simply because it works.
Apps that handle personal information, financial transactions or business operations still need careful testing and human oversight, regardless of who – or what – wrote the code. In one security study, about 45% of AI-generated code samples failed standard security tests.
The broader issue goes beyond coding. Researchers have long warned about “automation bias” – the tendency to accept recommendations from automated systems without adequately checking them. As AI becomes better at solving complex tasks, it’s becoming easier to accept its answers without questioning them.
Software is just one area where this can have serious consequences, because mistakes may remain hidden until something goes wrong.
AI can write code, but it can’t take responsibility
Vibe coding is opening software development to people with great ideas but little programming experience.
But making software easier to create doesn’t make it easier to judge whether it’s secure, reliable and ready for others to use.
No matter how good AI becomes at generating code, people – not AI – must remain responsible for what that code does.
- is a Lecturer, School of Engineering and Technology, CQUniversity Australia
- is a Professor, Information and Communications Technology, CQUniversity Australia
- This article first appeared in The Conversation
