I like writing code

AI Agents are ubiquitous in programming. Interviewing and assessment for programmer jobs now include how well you can order around an agent to do your job for you. My employer jumped fully on board and has been pushing something AINE, aka AI Native Engineering. I had a very awkward conversation with a higher-up about a project where using Cursor was mandatory. My disdain for that requirement definitely didn't go over well.

In programming, AI turns out to be pretty useful, despite being useless or harmful in pretty much every other sector. Code is about patterns and syntax, something a large language model can certainly regurgitate when you need to recall one of those patterns. While writing code, you'll need to re-learn things all the time, and it can be time-saver to describe what you want to an agent rather than go through the hard work of actually re-learning a thing.

The criticisms to AI are vast. It devalues labor, especially knowledge work. All models are trained on copyrighted work without permission. It pushes experts out of their fields and advertises itself as a replacement. It is an outright attack on creative work and art. It is an environmental catastrophe, given the amount of energy, water, and land required to build the data centers needed for compute power the models need. AI companies also destroyed supply chains and damaged personal computing to a degree that it may never recover by buying out essential components.

I might sound crazy when I say this, but AI is a fascist project. Control over information, how it's presented, and how you can spin it is all essential to propaganda necessary to prop up fascism. You may not have noticed this yet, but Google is no longer a search engine. It's an LLM prompt. When Google controls the model that people consider the easiest way to get to the truth ("just google it" is still a considered a wise thing people say to one another) then they can manipulate that model to achieve their political ends.

A core piece of fascism is to create its own fantasy that departs from reality. It doesn't surprise me that Hegseth's department of defense is obsessed with Claude for creating their plans. A sycophantic LLM that will indulge your fascist fantasy of the USA being an all-powerful juggernaut of war sounds almost too on-the-nose.

But given all that, I find that I don't like using an AI Agent while working is much more simple. I like writing code. Basically, I've seen a lot colleagues abandon writing the code themselves and adopt frameworks like SDD (Spec-Driven Development) in which the AI is writing 90-100% of the code.

I've tried these things, not only to see for myself, but also because it was required training for my job. I absolutely see the appeal. You can look really good to your bosses. You can also look like you are writing the code of ten developers if you tried to.

I was always an AI skeptic, so I can't say that I went in unbiased. But I do feel like it was a bit too easy. And I mean that in a bad way. When it's easy to make code, it's easy to make shit code. And I still believe code quality matters. Although, who knows. Maybe all code from here will be vibe-coded, and all future debugging and refactoring will also be vibe-coded, and then you'll never need to program "by hand" again.

A revealing thing a manager said to me about using an AI was, "it's like having someone with all the knowledge and none of the experience." Another buddy of mine said, "It's like having a really fast, overeager junior developer." Both of these situations sound like hell to me. I don't desire to be an empty vessel for an AI that I have to "get in line" like a babysitter. Nor do I want to manage juniors. I could have worked towards a engineering manager position at any time if I desired it. I do not. I actually like writing code.

The same thing happens to all engineering managers: their experience gets stale. If you aren't in trenches and writing the code yourself, your skills atrophy. Every engineering manager eventually becomes washed and can no longer offer useful technical input to a project. Sometimes it takes a couple years or a decade, but it always happens. This is what I think will happen to people who rely on an AI to do their work for them. They are not learning or exercising their skill anymore.

So through this whole saga, I'm waiting it out. I'm going to keep doing what I do, and I'm not going to sacrifice my cognition to a bot. Not only because I think it's best, but because I actually like working on the problems myself.