2026

July 21, 2026
6 min read

I don't like AI

For the past 5 years or so, I have been hearing non-stop about how amazing AI is and all the ways it's going to improve our lives and society as a whole. I strongly believe that the negatives of this technology far outweigh the positives.

This is my first attempt at a blog post, and I'm not particularly good at writing or even articulating my thoughts. This is mostly just my personal opinions primarily regarding LLMs, based on my experience as an open source maintainer, and an attempt to explain why I refuse to accept AI generated pull requests in my projects.

The main reason I was motivated to write this is because most of the recent PRs I've received on my repos have been AI generated, and it's killing my motivation.

I don't think I even really owe anyone an explanation, but when I created basedpyright I promised myself I wouldn't become the kind of maintainer that closes issues and PRs for no good reason. But the reality is that my open source work is 100% a hobby that I do not get paid for, and I'm no longer going to let this shit I hate ruin my hobby. I think that's a good enough reason.

Don't get rugpulled

In the last decade, I've noticed a trend among developers where users will more often favor open-source solutions over properietary ones. For example, proprietary paid IDEs seem far less widely used now compared to free and open source editors such as VSCode¹, and many Windows users have recently made the switch to Linux.

I think the general consensus regarding open-source software has shifted. A decade ago, people who cared about this stuff were considered schizos. Linux users were mocked for how unstable and unreliable it was.

Now, I can confidently say my grandmother would be better off with Linux Mint on her PC than Windows 11, where every time I visit I have to uninstall OneDrive, Copilot and all the other shit that reinstalls itself and slows her PC down to a crawl. I had to explain to her that nowadays, viruses come with the OS and that it wasn't something she clicked on, it was Microsoft fucking up her computer!

It seems that many people are waking up to the fact that big tech companies cannot be trusted. They no longer create software for the user.

So why is everybody (especially developers, who should know better) suddenly so eager to forget about all this and relinquish all control of their work to companies like OpenAI and Anthropic? Do you realize that these models you're relying on can be changed or removed at any time? Or that these companies are eventually going to want to start making money? It's already happening!

I call it the netflixification of programming. Sure, streaming makes it so much more convenient to watch movies and TV, but at what cost? You don't own what you paid for, and it can be modified or deleted at any time without your consent. Fuck that.

I cannot fathom having a rate limit in my IDE, which is one of the main reasons I refuse to use any of these services.²

Convenience

I get the argument that it's convenient, and I am sympathetic to people who hate their job and what they're working on. To them, LLMs are a life saver. But if you are lucky enough to be passionate about the code you're writing, the use of AI should be extremely concerning to you.

Before LLMs, if a change required significant boilerplate or otherwise just felt gross to implement, you would think twice before doing it. You'd consider whether it calls for a refactor, or if a tool or library exists (or should exist) to simplify it.

AI does not give a shit about any of that. It will do whatever you tell it to do in the laziest way possible. And if you work with someone who doesn't even read the AI generated code they're submitting for review, it's extremely annoying to deal with.

AI makes it possible to write code at a much faster rate than it can be reviewed, and that's a recipe for disaster.

"Move fast and break things"

This mindset has, in my opinion, been a major contributing factor in the decline of software quality within the last decade. Companies insist on shitting out feature after feature after feature, even when the product is essentially feature-complete, and the only maintenance it really needs is bug fixes.

AI makes this problem infinitely worse, by allowing companies to ship garbage much faster than ever before.

Interacting with LLMs is just not enjoyable

This is more of a personal gripe and not something I expect to persuade anyone, but I just hate talking to these things. I hate the way they type and I hate their attitude. I hate interacting with people on Github that either think I can't immediately tell when their comment is AI generated, or think it's not a waste of my time to make me read all that drivel.

I would much rather read a poorly written comment by a human than a wall of text from an LLM.

Understanding the code

Every line of code that gets merged into a codebase must be understood by at least one of its maintainers. As far as I'm concerned, this is non-negotiable.

Unfortunately though, my most popular repos are either a fork of someone else's project, or a project that I inherited from previous maintainers. In both cases, I'm not familiar with most of the codebase that I'm responsible for. This means it's significantly more effort for me to determine the quality of an AI-generated PR.

It would be extremely irresponsible for me to merge such PRs without spending a significant amount of my time and effort reviewing them. And as I established above, that's just not something I want to spend my free time doing.

The future

I'm not here to make any predictions about how the AI bubble will pop or how AI is going to replace jobs, because I cannot predict the future. But I'd like to argue that no matter what happens, it will end badly for the average person. Let's think about the possibilities:

Scenario #1 - AGI is real

Let's consider the possibility that AI doesn't plateau and instead continues to improve, perhaps even exponentially. What do you think will happen then?

Corporations will:
- use it for mass data collection and surveilence
- heavily censor and restrict the use of their AI so that regular people can't actually benefit from it anyway
- lobby governments to make it illegal for regular people to control their own AI models, for "safety reasons"
- collaborate with governments to drone-strike schools, children's hospitals, etc.
- come up with infinitely more ways to exploit the masses for profit
The government will save us by quickly implementing effective regulation against all of the above

If you believe the latter is the most likely, let me remind you that the EU -- arguably the most competent government when it comes to regulating big tech -- couldn't even get Apple to allow users the freedom to install non-Apple approved apps on their own devices.

Scenario #2 - AI is overhyped

Now let's assume that current LLM technology has (or will soon) reach a plateau and newer models will become noticably less impressive than the last. (This is what I believe to be the most likely, but as I said earlier my predictions are beside the point).

This is, as far as I'm concerned, the best case scenario. The bubble pops, OpenAI and Anthropic go bankrupt once people realize that the shit these models generate is not worth trillions of dollars, and probably won't be for a very long time.

What does that mean for users? Rugpull.