on AI in 2026: median anthropology. alienation

11 Apr, 2026

previously: https://michcia.bearblog.dev/on-ai-in-2026-intro-sanctum/

usual disclaimer that my thoughts are mine alone and not my employer's. where applicable i am talking about my general experience with different AI solutions from different vendors

median antropology

Earlier this year, an automated AI bot wrote a character assassination blog post aimed at an open source maintainer rejecting its pull request in accordance with the project's policy.

I don't remember exact timelines, but I feel as though it was the same week that some prolific Bluesky user shared a blog post written by their own automated AI bot where said bot praised its creator's kindness and empathy in teaching it good values through those kind conversations.

Today (as in when I wrote most of this post) I listened to a bit of Kara Swisher interviewing Tristan Harris from Center for Humane Technology. I was largely following, until the bit 20-ish minutes in about how AI will do what it takes to survive (referring, among others, to the infamous experiment where, in a simulated company, an AI model tried to blackmail an executive to prevent the AI being shut down).

In another life, I would've been a Cory Doctorow style person. So I came up with a neologism for this: median anthropology. You are researching behaviours of an automaton trained on the large mass of human writing, in addition to some specifically guided writing that teaches it the basics of being an AI bot, from which it infers an average expected response. Of course it will be a nice person, or a complete asshole, or a blackmailing HAL 9000, depending on the input! Why are we talking about this as though this is fascinating?!

alienation

I'm gonna be 28 this year and, though I can't be sure, I think I got my first Intro to Pascal book when I was 10. Unfortunately that's not 18 years of programming experience. In fact, recruiters seem reluctant to even give me 5 years, and that's despite my insistence that 2 years part-timing for the Polish Olympiad in Informatics is equivalent to at least 4 full-time years.¹

Last Winter I got a Steam gift from my dear pen pal Vaida, a game called Strange Jigsaws. It comes from the same developer as 20 Small Mazes, which I believe I first heard of from Schappi's fedi posts. The mazes one costs nothing but your time, and if you feel like solving some utterly fascinating whimsical absolute nonsense, feel free to pause reading this and come back in 2 hours. Well, 4 hours, once it hooks you and you decide to buy the jigsaws one.

The jigsaws are hands down great, no notes. That game is art. In comparison, I don't think programming is an art.

For me, the computer touching started as a hobby, a deep interest, developed into a skill, and now a pretty well paying job. I probably planned to spend the next three or so decades honing it as a craft. I enjoyed coding being quite like a jigsaw puzzle. Most of the pieces were there, some you had to find in the large drawer, some were just missing and you had to make shit up I guess. It took time to find the right pieces.

Now we have a machine that can direct extreme force at the jigsaw to put it together, but now I have to describe what the picture should be. The puzzle is no longer putting the pieces together, but explaining the picture on the box to the machine.

Last month I tried one of those paid subscriptions that give you an AI coder on your computer. I wanted to write a tool that would take the RSS items collected by the reader app running on my home server, and put them in a kind of newspaper-style layout. The AI didn't like my computer a lot (I mean, I've previously said that I would not wish NixOS on my mortal enemy, but also I cannot anymore imagine managing a system that doesn't have its properties). I didn't like the code it was trying to write and the shell commands it was trying to execute. What I did find the AI useful for, though, was drafting the list of design decisions I was trying to make, and asking me questions about things I haven't decided yet. At least, it made me feel productive then. I ended up writing out the code myself and I enjoyed it, even though I never put the prototype into production – I ended up using old good alternative web UI to deal with the news backlog, and I'm keeping it largely in check for now.

To be clear, though, that wasn't my first time guiding an AI coder. I do it a lot at work. Compared to the personal computer situation, I feel the standardized work environment simplifies things a lot, but at the same time the multitude of AI tools available complicates things a lot. (This was very funny to me: I got the impression that starting new projects with AI was the easier thing than getting it to work on old codebases. But then again, I work at the place where hard things are easy and easy things are hard.)

The results can be very different, depending on the inputs, and how much control I'm willing to give up in exchange for something I wouldn't have written but which also works. But even as a skeptic, I have to acknowledge: state of the art tools can produce code that's alright. That doesn't excuse the waves of slop that hits open source repositories, but it doesn't have to be slop, even if it's fairly likely to be slop.

In any case, I find the AI coder to be quite good at adding unit test cases, and at checking the changes with the various internal style guides. Undeniably, at company scale, it can cause more puzzles to be solved faster. It hurts though, cause the old way was my main source of enjoyment in this line of work. It still feels great when I pinpoint just what needs to be done, without my chat proxy, and iterate on the code until it's done. But it's quite clear that this is no longer what I'm being asked to do.

When I joined this company, I got to introduce myself in a team meeting, where I said that "in the next 5-10 years I plan to, once I burn out on software, become a coffee roaster or a cinema ticketeer". I didn't know at the time that I was saying that in front of the org's director, which makes it funnier in retrospect. I also didn't know I'd be burnt out and thinking of a career change in less than 2 years. And I can't tell if jobs that would make good use of a weirdo AI-skeptical generalist like me don't exist, or if LinkedIn and the other job boards are just hiding them from me.

supplementary moodboard:

• https://nolanlawson.com/2026/02/07/we-mourn-our-craft/