How I Develop with AI in November 2025
Let me start a technical debate for developers.
I pay $10/month for GitHub Copilot. I have colleagues spending over $100 on Cursor and Claude Code. And although they constantly ask me if I should switch, I stick with my setup.
Not because it’s the best. Not because I can’t optimize. But because I need to balance productivity with control, and for now, this balance works for me.
Let me tell you how I work with Copilot, what limitations I have, and why I believe price isn’t everything.
How I Use GitHub Copilot (Integrated in VSCode)
What I Love
1. Native Integration
- I can highlight code and apply actions super easily
- I switch between agent mode and normal chat depending on whether I need changes or just understanding
- No tool switching, no friction
- I can add context of certain files also quite easy
2. Adaptation to My Style
- Even though I don’t have a documented style README, it follows my programming approach quite well
- It adapts to the context of the project I have open
- I don’t need to explain my architecture every time
3. Single-Repository Focus
- I work with about 15 different repos, but I handle them individually
- When I need to touch multiple repos, I do it one by one
- I prefer having control over what changes are made in each repository
The Unexpected Problem: Silent Errors
Here comes the interesting part: especially on the frontend, which is where I have the most code and perhaps review line by line the least, I’m finding unexpected errors.
They’re not obvious errors that break everything. They’re those subtle AI changes that go unnoticed until something doesn’t work as you expected.
This is my biggest current concern: delegating so much that I lose visibility into what’s really happening in my code.
GitHub Copilot Integrated in GitHub: Beyond the IDE
It’s not all about writing code. Copilot saves me in two key moments:
1. During Deploys
It has saved me more than once. When I deploy and there’s an error, I can ask it directly in GitHub what’s failing without having to change context.
2. In Commits
It makes automatic summaries of my commits. Seems trivial, but when I need to review what I changed two weeks ago, these summaries are gold.
The Reality: I Still Code
I’ve heard several colleagues say: “I’ve stopped programming”.
Not me. I still code.
What I barely do is type code like before at 100%, but:
- ✅ I review the generated code
- ✅ I edit AI mistakes
- ✅ I improve code to make it more understandable
- ✅ I think about where and how to place things
- ✅ I add my practices and criteria
In an advanced project, you CAN'T NOT code. Maybe in a brand new project from scratch you can delegate more, but in one with complex logic you have to adapt, review, and make sure you don’t break anything.
My Limitations with GitHub Copilot
The Credits Problem
I pay $10 and sometimes I run out of credits to use the more advanced models in the last days of the month.
I notice it. The basic models are noticeably “dumber”.
The next plan is $40/month. I’d be paying $30 more, but for now, I prefer to save them while I have things to launch.
The Real Limitation: My Brain
This isn’t a Copilot limitation, but I’m not able to develop things in parallel.
The agents don’t take so long that I can leave a task to review another. My head explodes trying to track multiple contexts.
Because of this, I don’t feel I can optimize my work much more thanks to AI. How much could I improve? 5%?
That’s why I don’t usually dedicate time to it for now while I’m busy with the product.
The Context: I’m Not Just a Developer
I know there are people with super optimized AI development workflows, but my reality is different: I don’t spend all day glued to the code.
I have to balance between:
- Support
- Talking to clients
- Thinking about business
- Product strategy
- Technical decisions
I know there are people who don’t sleep and do a thousand things, but that’s not my case and it won’t be. For me, coding is a way to solve X things, not my main job where I need to optimize everything to the maximum.
If I were a full-time developer, maybe I’d invest more time in perfecting my AI workflow. But as CPTO, code is one tool among many, not the only one.
Why Don’t I Use Other Tools?
Claude Code (Cline)
For my taste, accessing from the console is a pain. I haven’t tried it in depth, but that’s the feeling I get from the start.
If I already have a fluid integration in VSCode, why add friction?
Cursor
I tried it once. The problem: changing IDEs and having to clone all my repos made me lazy.
My free trial ran out (which didn’t last long) and I’ve never tried it again. Inertia is powerful.
What I Know I Could Optimize
If I had time (spoiler: I don’t right now), this is what I would do:
- Better document each repository so the AI understands it better
- Create a README of my programming style that I can reference
- Use multi-repository functions to work with multiple projects simultaneously
But the reality is: for now I prefer not to lose focus while we have a product to launch.
My Conclusion: It’s Not Just About the Price
Could I have more features for $100/month? Probably.
Would it make me 10x more productive? I doubt it.
For me, the current balance is:
- ✅ Low cost ($10/month)
- ✅ Frictionless integration (native VSCode)
- ✅ Enough power for my workflow
- ⚠️ Sufficient control to not lose visibility
I'm not looking for the perfect tool. I'm looking for the one that lets me keep building without adding unnecessary complexity.
What About You?
This is everything I have to say about how, as of November 5, 2025, I’m using GitHub Copilot.
Do you use other tools? Have you found the balance between delegation and control? Do you think I’m leaving productivity on the table?
I’d love to hear your experience. Especially if you’ve tried Cursor or Claude Code and have real production feedback.
This post reflects my personal experience. I’m not sponsored by anyone (I wish 😅). I just want to open a real debate about tools we use day to day.