Vibe coding is all the rage today. When you think of it, you might picture creating full-stack apps, sleek websites, or trendy portfolio sites. But for me, vibe coding began with a simple idea when OpenAI released ChatGPT.
My first project? A Chrome extension, not a website or app. I’ve had many ideas bouncing around, but I wanted to test three major LLMs (Claude, ChatGPT, and Gemini) to see how they would create a Chrome extension from scratch.
Chrome Extensions: A Simple Test
Chrome extensions are perfect for vibe coding. Unlike full apps or websites, you don’t need to worry about hosting or databases. You simply load your files into the Chrome Extensions section, and you’re good to go. Each extension usually consists of a manifest file and some code.
I chose an idea that felt useful: a tool to search through Instagram story viewers. If you’ve ever tried to find someone among a sea of names, you know the struggle. I gave each model a vague prompt and let them run with it.
ChatGPT’s Effort
I started with ChatGPT. It created a ZIP file for the extension in just a couple of minutes. Unfortunately, when I tried to use it, it had a few serious flaws. The initial version lacked a popup, so I couldn’t even click it. After some fixes, I got a search bar that mostly worked—but only for viewers on the screen. Its methods were a bit clunky, which made me doubt its practicality.
Gemini’s Approach
Next up was Gemini. I was excited because I’ve enjoyed its capabilities before, but I quickly got disappointed. Instead of providing a functional extension immediately, it gave me steps on how to create files on my own, which was confusing. Even after following its instructions, the extension failed to work correctly. After multiple tries, it became clear that I needed to give up.
Claude Delivers
Finally, I tested Claude. Its initial output didn’t work, but with a quick redevelopment relying on Instagram’s API, it delivered a functional extension. Claude created a search bar that worked perfectly, allowing me to find viewers on individual stories. Within just a few messages, I had a tool that was both practical and user-friendly.
Conclusion
In the end, Claude stood out as the champion of the three models. It produced the best result with less back-and-forth. As a bonus insight, LLMs like Claude could streamline digital tools even further, as seen in a recent survey by TechCrunch where about 70% of developers expressed interest in AI-assisted programming. Next time, I’ll start with Claude right from the beginning.
For more resources on vibe coding, you can explore the OpenAI website for cutting-edge technology and tools.
