No AI was used to write this article, therefore might have some English mistakes. Also, as many articles I wrote, this one will be for a quick copy-and-paste link when someone ask me about it ha!

This post doesn’t reflect the opinion of the company I work for (it’s kinda of an obvious statement but sometimes you need to say it, because people don’t understand that folks have different opinions), in fact this subject has been giving fun occasional discussions in the lunch time.

Friends, coworkers and folks from the tech community keep asking me about if I still write code, yes I do.

Lemme go over a quick background story first…

I really like programming, if you follow my work or have worked with me, you know how passionate about software engineering I am. I don’t want to enter on this point, neither saying things like “hey look at me spending weekends to code” or happy hours of discussion to debate how a terminal emulator works, let’s just stablish the fact that I like coding.

I also actually like AI, more specifically language models. I find it an interesting tecnology and I do read 1-2 papers about AI (occasionally in the subset spectrum like ML, DL and etc) every week in my kindle on the way to work (almost 1h train fwiw). Love the mathematical engine behind it and I like how you can debate models architectures with AI researchers for different task purposes leading to pros and cons.

That being said, when LLMs came to the picture for coding work. I was very reluctant and well, very against using it. To be honest, I only really started to use AI for coding when I started at my current gig this year (2026) and for the record it’s not because I am working with AI (I was working with it before in my previous work and I was completely against it).

Truth is that models are getting better in writing/analysing code and they will get better, it’s time to accept this fact and move on. I made peace with it.

However being faster in write code doesn’t necessary means that the output is good and scalable. Since people are starting to become 100% dependant on models to do any kind of work in the software field, it’s actually becoming a problem.

Probably you have heard about the term “meat proxy”, if not, it means “a person who copies and pastes AI-generated text, code, or messages and sends them to others without reading, checking, or understanding the content”. I think this term isn’t used enough, and it should be even more used for engineering IMHO.

More often I see people building software that’s completely unrealiable and the same people, are losing the skills to decide what is good or bad. Most of the developers lately aren’t deciding anything but in fact relying on models to make decisions for them. I decided I don’t want to be part of this cycle, I want to keep studying, to keep being a “sharp knife”. So, this is what I have been doing: Agentic work by day, and in my free time coding by hand (I have not idea what’s the right term for it nowadays).

Specifically developing a project bigger than the terminal emulator I did 4 years ago, but now using my own programming language instead of Rust. I’ve uninstalled all AI stuff in my personal machine and have been using my own programming language for coding. To be honest, this setup has been kinda of “anti-AI” because the models 1. doesn’t know my language and 2. there’s zero context on the ecosystem.

So far it has been working for me, I have the best of both worlds and I am happy with it.

BTW it’s really funny when I open an editor to edit text these days, people getting shocked by it. This always makes me laugh.

Best,

Raphael.