The problem with "vibe coding"

By Aatu Väisänen

April 25th, 2025

In the last couple of months, the software development space has been introduced to a new trending term called “vibe coding.” Originally coined by Andrej Karpathy (one of the founders of OpenAI) in a popular X post, vibe coding refers to a new kind of way to create software with the help of LLMs (large language models) and agentic AI tools such as Cursor and GitHub Copilot. You just keep prompting the AI to generate code and keep iterating until you are satisfied with the result. The important part is that you don’t care about code quality, possible security issues, or other ethical concerns – you just give yourself fully to the “vibes” and accept any changes suggested by the AI. Heck, vibe coding might not even require you to know much about programming, as you are only giving natural language instructions.

Does vibe coding equal AI-assisted software development in general?

Technically, vibe coding requires the use of the same tools that any AI-assisted software developer would use: LLMs, and more specifically agentic AI. The difference is that if you’re vibe coding, you simply don’t review the code generated by AI. A responsible software developer leveraging AI tools in their daily work takes the time and effort to review and adjust any code generated by the AI. In many cases, the AI does a good enough job that doesn’t require exhaustive editing, but even so, it should not be understated how important it is to always review the AI’s code output. This becomes especially important when creating complex, business-critical logic.

So… what’s the problem with vibe coding?

With the quick advances in AI coding tools, creating functioning software is becoming more accessible than ever before. In theory, anyone with access to tools such as Cursor is able to prompt their way into creating a fully functional application in a matter of hours or even minutes. Just a few years ago, only a really experienced software developer could have completed such a project in a short time.

With more and more people pushing out vibe-coded applications, things like security vulnerabilities and biases embedded in the AI-generated code become a widespread problem. Software created with the vibe coding mindset also lacks proper testing, which leads to more user errors. On top of this, the code becomes hard to maintain or scale, creates inconsistencies, and lacks coherent structure.

Should you vibe code?

The issues with vibe coding can pile up and cause massive technical debt. But does this mean it should be avoided at all costs in every scenario? I would say... no. You should definitely give vibe coding a try, but only for certain purposes like prototyping, experimenting, and hobby projects. Just DON’T push “vibe code” to production in any professional software project!