How I used AI to build a Chrome extension

Here’s everything I learned using Google AI Studio to build my first Chrome extension.

Give me the TL;DR

After two years of dreaming about building my ideal Chrome extension for SEO, I finally built a working prototype in two hours for free using AI.

Which AI tool(s) did you use?

For 100% of this project, I used Google AI Studio. I mostly used Gemini 2.5 Pro, but when Gemini 3 Pro Preview came out, I switched to that.

What was the best part about using AI to write the extension?

While I can write code in Julia and other languages, I never really learned to write Javascript, the language Chrome extensions are written in.

I’ve also never written a Chrome extension before.

And with life being busy, this was very much a case of “use AI or it simply won’t get done”. So getting it done was the best part.

What didn’t work well?

So there’s one feature I really wanted to add to the extension but the AI just couldn’t do it. The feature was the ability to have the panel automatically close (if open) on selected hostnames (domain names). For example, we use Jira and Confluence at work, and I wanted to have the extension automatically close when on those pages (but remain open for non-Atlassian URLs). The plan was to make the hostnames user-configurable, but that was my use case.

I know the auto-close functionality is possible, but the AI just couldn’t do it—probably due to a hidden clash with other logic.

I tried a few times with Gemini 2.5 Pro, and then when Gemini 3 Pro Preview came out, I thought that might get it working for me. I’ll try again with future versions.

Apart from that, there were many times the AI confidently told me it had successfully implemented one of my feature requests, only for me to find it was flat out wrong, and that the new feature didn’t work at all. I was able to eventually implement all these features—just not the auto-close functionality.

How much did you spend on this?

Nothing in dollar terms.

How long did it take to make?

About 20-30 hours (over a few weekends) of going back and forth with the AI.

Most of this was due to me adding features, but about 25% or less of this time was due to the AI getting it wrong.

At this stage, there’s no ability to have Google AI Studio check its own work, so I had to reupload the files as a local extension (“Load unpacked”) at least 100 times. This is the “human as bridge” stage of AI.

What did you learn?

A few things:

1. You can edit the files yourself within Google AI Studio. This might sound obvious to some, but I spent ages asking Gemini to update simple things such as tooltips in the extension, when it would have been quicker—and better for the environment—for me to do this myself. Lesson learned.

2. If you ask it for ideas, be sure to tell it not to implement any changes (unless that’s how you roll). Once, I asked Gemini what it thought would make the extension better. Instead of simply telling me its idea (add AI-powered SEO recommendations), it tried to build it without even seeing if I wanted that. I didn’t. It didn’t work anyway. I reverted the changes using the revert-to-checkpoint feature within Google AI Studio.

3. For me anyway, Google AI Studio created a (blank) file called index.tsx. I couldn’t delete it within the interface for some reason, even though I did have the options to delete other files manually. Long story short, the Chrome Web Store wouldn’t accept my extension until I deleted this file.

Final thoughts?

It’s incredible to see how fast I built this extension despite not being a Javascript coder. I recommend giving it a go, even if the app or extension you want is just for you!