Vibe Coding: Beyond the Buzz

The Origin Story...

Vibe Coding has become one of the latest buzzy topics in tech. Originally coined by Andrej Karpathy on X, one of the co-founders of OpenAI, to describe the Zen-like feeling of abandoning the details of your code to your AI assistant. Does it work? Who cares? Just ask the AI to try again until it does.

If that thought makes your blood turn to ice, keep in mind that Karpathy's original post was about "...throwaway weekend projects." If you are someone that has fully embraced the help of one of the myriad of AI coding tools, such as GitHub Copilot, Google Gemini, or a full AI IDE like Cursor, you're probably already familiar with this style of coding, even if you didn't know it had a name.

Practical Applications?

From our experience it becomes really easy to trust the code written magically for you. For our own "weekend projects" it becomes second nature to just describe what we want and just accept what the AI outputs. Our developers find that they end up focusing more on architectural decisions and what they actually wanted the code to accomplish rather than how to implement it. The result was a whole project's worth of code very quickly, but there was one problem. Did it even work?

Well, no, it didn't. That's where years of experience working on new and unfamiliar projects started to come in handy. With a little bit of debugging a little demo app was functioning. Some of the issues you'll run into with AI generated code is that it only knows what it was trained on, so you may have to fix a lot of deprecation warnings. The code can also just be messy, with lots of duplication or code that does a lot of work setting up data that never gets used.

Link to a funny YouTube video on this topic.

Conclusion

So Vibe Coding is great for getting past an empty screen, or experimenting with new technologies, but eventually the vibe has to shift back to more traditional coding practices. For existing projects, an AI assistant can still help, but its use would be more targeted and much smaller in scope. The decision of when to use AI will, in the end, come down to personal choice.