• 22 Nov, 2025 *

The AI bros might have been on to something

I’m not necessarily proud to post this but it’s just a reality of where I’ve been the past couple of months: nearly all of the code that I push is generated by AI. When I think, I prompt. When I run into friction, I prompt. The act of typing out code by hand was a pleasure that I thought would never leave me. Yet here I stand. A man who knows the latest llm models by name, their pros and cons, and begging my company for more token credits. I have taken my brain and effectively handed it off to OpenAI (well, technically anthropic because sonnet 4.5 is pretty sweet).

There are pros and cons to using an llm this much but I’m not necessarily interested in talking about those as you can find a billion articles on the topic. But you know what you can’t find? An article written by ME about my descent into brave new world. How did we even get here? Is there any hope for a way back? Do I even want things to go back to normal?

We got here but... how?

So, I work for a very small company. There’s only 3 devs, including myself, and every day feels important. We don’t have infinite money (the opposite in fact!) and we are constantly building something from 0 to 1. Needless to say, we are a startup delivering things at breakneck speed. Ok, so that’s part 1. Part 2 is my CTO telling me “have you used cursor before?”

Up until I joined this company, I had been using nvim with a somewhat janky copilot plugin. Good times. I could ask copilot questions about my code, map out obscure codepaths, and if I was feeling crazy, have it generate some tests for me! Wow, 2024 was awesome! Fast forward to my first day on the job. A fresh new laptop, infinite possibilities. Why not give cursor a spin? My plan was simple: write most of my code in nvim and then use cursor as a “fill in the blanks” type of solution. And for a while, that’s what I would do. But then the pressure of delivering set in, moving fast was no longer a stretch goal. I needed to strap a rocket to my computer and write code as quickly as I could.

My initial “wow” moment using cursor was being able to describe various frontend components and the llm spitting out nearly exactly what I wanted in 30 seconds. This was awesome because I have hardly any interest in the frontend and you’re telling me I can just automate that? Yeah, I’m in. Since all of our features involve some kind of UI counterpart, I would just write my backend code and have the llm do the boring part. And since js/ts is sooo ubiquitous, it was easy to trust that the llm was answering the question correctly. From there, it’s not hard to see how I then eventually moved to having it write backend code for me. I just needed to add some context to the prompt and it knows what to do.

And we are now caught up. All of my code is basically generated via a prompt. I used to laugh at CEO’s who made claims about how much our code would be generated by AI. But now I find myself the subject of one of their case studies.

My new home

This feels like my brave new world and it feels worth it to reflect on whether or not I want to continue down this path. So, a couple of questions.

Why did I even fall in love with programming? It was because I loved making computers talk. To be able to type some input in and get output out... I know it’s simple but I love it! And while using an LLM does offload some of the intellectual work, I still get that dopamine hit when I see a feature come to life. But getting things to appear on a screen isn’t the only thing I like about programming, it’s also that writing software is an incredibly efficient way to help people solve problems. Even if it isn’t necessarily a “deep” problem I’m solving, helping people by using computers is rewarding.

Another question: do I want things to go back to how they were? Not really. I don’t mind that I’m not really writing code anymore. I like that I can set that task aside to the LLM, make some follow-up comments, and move quickly. I’m churning out features faster than I ever could have imagined and (see above) that’s fun! It’s rewarding! BUT, what I do worry about is becoming stagnant and not letting my brain work on hard problems because the answer is just a couple of prompts away. I’m worried that I’m tricking myself that I’m writing quality code because it Just Works™️. I don’t want to trade in my passion and intellect for performance gains.

Where to next

Ok so that last section was partly a lie. I do miss typing out code by hand but I can’t really justify it at the stage my startup is at. I need to be moving as fast as possible and llm’s are just the best tool for the job right now. But I do want to make sure that I’m finding a solid amount of time to get some side projects in that serve absolutely no purpose except for me to learn. And I want to make sure that I’m not using an llm to write out the code for me. I’m going to focus on getting some side projects done in the new year that are enjoyable and interesting to me. Projects that don’t have a long term vision, perfect for a weekend.

At the end of the day, I don’t have a problem that I use llm’s extensively now. It’s not the future I wanted to be a part of but it’s made me a better co-woker and has been good for the business. My passion for computers will always be there, I’m very sure of that. I just want to make sure ai tools aren’t a way to speedrun through my assignments with no thought of the consequences that could come my way. Vibecoding should be avoided at all costs, at least professionally. As for me, I think those are habits that I need to fight against as the temptation to be lazy comes a litte too naturally to me.

Life can be a little too easy sometimes. Tread with caution.