This content originally appeared on DEV Community and was authored by Ankitkumar Singh
Few months back, I caught myself doing something ridiculous. I was copying the same project update into different systems – our timesheet portal and the client project management tool. Same text, different formats, 20-30 minutes gone. The worst part? I’d been doing this every single month.
That’s when it hit me. I’m a developer. Why am I living like this? And yeah, then I decided to spend the next 3 days of holidays (Saturdays and Sundays) automating this task so that I’d be able to save around 20 minutes on average.
The Timesheet Problem That Started Everything
Our company timesheet system is… let’s call it vintage. You click into a cell, type your hours, add a description, select a project code from a dropdown that loads like we’re living in the 90’s era, then repeat for every single month. If you work on multiple projects? Multiply that pain.
I started simple. Really simple, actually. My first Chrome extension just filled in “8 hours”, selected the project from the dropdown, and added an empty description. That’s it. But even that tiny automation saved me thirty seconds, and more importantly, it proved something: I could bend the browser to my will. Yeah, it felt so good that I could build something to solve my own problem.
Within a week, I’d expanded it. The extension now:
- Auto-filled common project codes based on the day of the week
- Copied descriptions from yesterday with one click
- Batch-filled entire weeks for recurring tasks
What used to take five minutes now took twenty seconds. But here’s where things got interesting – I realized the pattern recognition I was doing manually could be handled by AI.
Enter the AI Models: My ChatGPT Prompt Automator
After the timesheet success, I noticed another pattern. Every morning, I’d open ChatGPT and type variations of the same prompts:
- “Review this PR description and suggest improvements”
- “Convert this JSON to TypeScript interfaces”
- “Explain this error message: [my text]”
- “Rewrite the sentence and make it formal: [My text]”
The prompts were similar, but I kept retyping them. So I built a Chrome extension that automated this. It automatically:
- Opens ChatGPT
- Shows prompts and on a click pastes everything
- Hits submit
You can check out the full journey of building this extension in my previous post: How to Build Chrome Extensions Using ChatGPT & JavaScript. That extension became the foundation for understanding how AI could help me build even more complex automations.
But wait, you might think – that’s just a fancy copy-paste tool. And you’d be half right. The magic came when I started using the AI models themselves to build better versions of the extension.
How AI Models Became My Coding Partners
Here’s what changed my entire approach: I stopped trying to code everything myself and started treating Claude, Gemini, and ChatGPT as specialized teammates.
Claude for Complex Logic
When I needed to parse our timesheet’s ancient HTML structure, I literally took a screenshot, uploaded it to Claude, and said: “Write a function that extracts all the input fields from this mess.”
Claude didn’t just write the code – it explained why the site’s structure was problematic and suggested defensive coding patterns for when it inevitably breaks. The function it generated handled edge cases I hadn’t even considered, like when the page partially loads or when sessions timeout.
ChatGPT for Rapid Prototyping
ChatGPT became my brainstorming partner. I’d describe what I wanted: “I need a Chrome extension that detects when I’m on a Jira ticket and automatically generates a Git branch name from the ticket title.”
What I got back wasn’t production-ready code, but it was a working prototype in minutes. The manifest.json, the content script structure, the basic logic – all there. I’d then refine it, but starting from 60% complete beats starting from zero.
Gemini for Documentation Parsing
Gemini surprised me. When I needed to understand Chrome Extension Manifest V3 (which, let’s be honest, has documentation that feels deliberately confusing), I fed it specific API docs and asked: “How do I make a content script that only runs after a button click?”
Not only did it explain the concept, but it also showed me the permission pitfalls I was about to walk into. Saved me hours of debugging “why isn’t this working” issues.
The Time Math That Made Me a Believer
Let’s talk real numbers. Between these extensions and a few others:
- Timesheet filling: 5 minutes → 30 seconds daily
- Repetitive ChatGPT prompts: 2 minutes → 10 seconds per use (probably 10 times daily)
- Context switching after meetings: 3-5 minutes → one click
- API response analysis: 2 minutes → instant
Conservative estimate? I’m saving 45 minutes every single day. That’s almost 4 hours per week of mind-numbing clicking eliminated. And remember, I spent just 3 days building these. The ROI hit positive within the first week.
Building Your First Automation
You want to start? Here’s exactly how I’d do it today, knowing what I know now:
Pick your most annoying daily task. Not the complex one, the annoying one. The thing you do every day that makes you think “there has to be a better way.”
Open Claude/ChatGPT/Gemini and describe it naturally: “I have to click through three menus to get to my timesheet every morning. How can I build a Chrome extension that takes me there directly?”
Start with the manifest.json they give you. Every Chrome extension needs one. The AI will generate a basic version that actually works.
Test immediately. Load it as an unpacked extension (chrome://extensions/, Developer mode on, Load unpacked). It probably won’t work perfectly. That’s fine.
Iterate with AI. Copy any error messages back to the AI. Be specific: “When I click the button, nothing happens. Console shows: ‘Cannot read property of undefined’.”
The thing is, you don’t need to be a Chrome extension expert. You need to be good at describing problems to AI and testing their solutions.
The Gotchas Nobody Tells You
Let me save you some pain. These are the things that tripped me up:
Manifest V3 is weird. Background scripts are now service workers. You can’t use XMLHttpRequest. Some perfectly logical things just… don’t work. When you hit these walls, explicitly tell the AI you’re using Manifest V3.
Permissions matter. The AI might suggest code that requires permissions you haven’t declared. Always check the manifest.json permissions against what your code is trying to do.
Content scripts are sandboxed. They can’t directly access your extension’s storage or make cross-origin requests. You’ll need message passing. This confused me for days until Claude explained the architecture.
Test on multiple sites. Your extension might work perfectly on one site and break on another because of Content Security Policies or different DOM structures.
Where This Gets Really Interesting
Here’s what I’m still working through: these AI models are getting better at understanding context. I recently described a complex workflow involving multiple sites, and Claude suggested a multi-part extension that:
- Detects which part of the workflow you’re in
- Automatically proceeds to the next step
- Maintains state across different domains
I didn’t even know that was possible with Chrome extensions. But it is, using a combination of content scripts, background workers, and the storage API.
The code Claude generated wasn’t perfect, but it introduced me to concepts I didn’t know existed. That’s the real value here – AI as a teacher that responds to exactly what you’re trying to build.
Start Tomorrow, Thank Yourself Next Week
Look, I’ll be upfront – your first extension will probably be janky. Mine definitely was. It only worked on Tuesdays (seriously, I had a date bug that took me ages to find). But even that broken Tuesday-only extension saved me time on Tuesdays.
The barrier to entry is lower than you think. You don’t need to understand the entire Chrome Extension API. You don’t need to be a JavaScript expert. You need to:
- Identify a repetitive task
- Describe it to an AI
- Test and iterate
Actually, let’s make this concrete. Tomorrow morning, pay attention to the first annoying repetitive thing you do in your browser. Write it down. Then spend your lunch break asking ChatGPT or Claude how to automate it. You might not finish the extension, but you’ll start it. And starting is everything.
The path from “I click these same five buttons every morning” to “my extension does it for me” is shorter than you think. The AI models are like having a senior developer who’s infinitely patient and available 24/7. They won’t judge you for not knowing what a service worker is. They’ll just explain it and help you use it.
My timesheet now fills itself. My prompts are a keystroke away. My browser remembers what I was doing before that meeting derailed me. These aren’t revolutionary innovations – they’re personal quality of life improvements that compound every single day.
Your browser is programmable. AI makes that programming accessible. The question isn’t whether you can build these automations. It’s whether you’ll start tomorrow or keep clicking those same buttons for another month.
Me? I’ve got 45 minutes back every day. What could you do with yours?
This content originally appeared on DEV Community and was authored by Ankitkumar Singh