Note: This isn’t a coherent post. I am not using any LLMs to punch it up, nor am I gonna do a second draft, because I think this would lessen the words I want to say.

It’s been 6 months since I’ve been unemployed, and let me tell you - does not feel good! It got so bad I even started writing in a journal sometimes (and because I’m a nerd, of course I had to use jrnl instead of a normal, written journal (and also because I write in the world’s worst chicken scratch)). Noted, this might be partially on me for being somewhat lassaiz-faire with my money (I really shouldn’t have bought that 3D printer), but hindsight is 20/20.

During that time, AI development progress went into turbo-overdrive, much to my chagrin. I’ve always been an AI skeptic, much to the dismay of my two previous employers. Thankfully, in one, the team lead understood my point of view and respected it. In another, however, I was not afforded this luxury, as the developers treated it as a replacement for documentation or just hopping on a call for them to explain something.

This moment further soured my view on AI, because talking to people is how I do my best work, cleaning up misunderstandings, learning from others, getting a second pair of eyes on my code to see if I blinkered myself too hard. I’ve always prided myself on that, bragging about that as my strong suit in every interview. And now… that sort of connection I’ve appreciated for my entire career is being replaced by roleplaying with a hallucinating spreadsheet.

Disdain of people

So, we all noticed it right? That a lot of the marketing around LLMs and genAI is terribly anti-other people? A lot of it comes from “solution in search of a problem” nature of it all, where the marketing is bending themselves over backwards trying to find a reason for spending $1.4T and not making a penny off that investment, but a lot also comes from, I believe, the leadership being so incredibly hateful of people that do not actively make them money.

Years of being forced to reply to menial emails, being stuck in meetings, why not zone out and let an AI take care of everything, while you knock off for a nice relaxing vacation? But you can’t put R&D effort into something that will only make some money. You need to make it everyone else’s problem, because The Line Must Always Go Up. You always have to chase the myth of perpetual growth.

So instead of asking your neighbor about how to do something, or (god forbid) paying someone a reasonable amount to fix something for you (or teach you how to fix it yourself), or giving an artist what their labor is worth, we’re told that the computers can do it all instead! “Isn’t technology wonderful?” they cry, as they build barriers and barbed wire around other people, telling you they’re to be feared or envied, rather than that they could be your friends. This, and the myth of a visionary, an exceptional individual, an entrepreneur; is all a ploy to make us hate our neighbor, view them as the enemy, which… Come to think of it that does sound like the usual sort of racism and xenophobia many of the elite are indulging in as we speak! Funny that!

The singularity is pathetic

I don’t have any more profound opinions on the nature of humanity in this section though, this one is just me expressing my frustrations with the entire user experience of LLMs. I reluctantly started trying out LLMs earlier this month, mostly just out of frustration with the job market. I justified it with thinking “if they’re gonna slop me, I’m gonna slop them right back”. Of course, me being me, I sure as hell wasn’t gonna give Anthropic or any other AI provider any money, so I decided on selfhosting it.

I popped Ollama and Open WebUI (which, by the way, is the world’s worst name for an open source project) in my NixOS config, downloaded Gemma 4, installed Claude Code, and to dip my toes in, I tried out career-ops, a project promising a way easier way of applying for jobs, with integrated CV generation. “Perfect”, I thought, “something that will help me out with this sisyphean task”. I set it up, answered all the questions it had, and was ready to start unleashing this little pest on the world of job descriptions. I used the skill that was said to analyze job descriptions, gave it a link to LinkedIn, and watched it go.

By this time, I already got how these agents worked, when they use an external tool, a little notification pops up that such a tool was used/in use. I waited a bit, watched it exchange tokens, and it proudly stated that it had called a tool and is waiting for it to return! But something was wrong, because I saw no notification like that.

I thought that maybe it decided to just do that in the background without any sort of feedback, lord knows a bunch of software that does that already exists, so I went to do something else. About 5 minutes later, I came back to it, and saw that it was still on the same screen as I had left it. I reran the same skill, and it told me that it’s still working on the last one. So I went and did some other things.

I come back, same thing. I thought “oh right, people talk to their agents”, so I asked the agent about it. It spat out that it is still working on it. I confronted it, saying that it didn’t call any tool, it kinda just ignored me. I told it that I saw nothing in the process list. At that point it did actually run the skill.

Relieved, I kept using it. However, the same problem came back, except this time, the agent spat out that… it gave up. It wasn’t gonna run the skill I wanted it to run. It just went “let’s do something else”. No, you deranged facsimily of Cleverbot, let’s not do something else.

At this point some of the more frothier vibe coding afficionados will pop out of the woodwork and say that I should’ve written a better prompt or used a better skill or used a better model. Yes, maybe I should’ve, but that is so far besides the point. If the radio in your Skoda Fabia breaks, you fix the radio. You don’t tell someone that they should buy a Jaguar, or even worse, rent it every single day.

Why the obsession with turning engineers into managers by making them write long specs in plain English and having them babysit a junior junior junior developer? Why should I work with a tool that can just refuse to work for no particular reason?

The thing I like most about code is that if I write something, it will mean that thing. It won’t mean anything else, it will mean that thing that I wrote. If I wrote it incorrectly, that’s on me. But bringing in the infinite tapestry of human interaction into this mess just creates even more ways for things to go wrong, and you don’t fix that by swapping to a better model, or by repeating constantly what you mean and don’t mean to the LLM.

Argument Simulator

See, this is what I do not understand with the higher-ups. Why are they so obsessed with trying to make me talk to every single of my devices? Both out loud and in a chat interface? Why can’t I just get a normal, intuitively laid out menu that I can grok (pun intended) really easily?

I thought a lot about this, why do companies think that instead of using my phone in silence, I’d want to talk to it out loud (and on speaker! Not with headphones!) and argue with it over what I should do? What’s faster: blabbing to a statistical model out loud in public, or just pressing a couple buttons?

Then I realized: it would only make sense if the people using it weren’t in public a lot of the time. It would make sense if they had plenty of opportunities to, say, drive a car a lot, or sit at home. Kind of like what Americans do. And Americans also find it somehow culturally appropriate to talk out loud on speaker phone in public.

Here in Poland, if you walk into a public space with music or, god forbid, TikTok, on speaker, people will look at you weird. Even worse if you’re talking to someone on speaker. Hell, they’d look at you weird if you were talking to someone over the phone. That is just the kind of society I live in, and I quite like it, actually. And I do not appreciate being forced into talking out loud to my electronics.

So please, can I please keep talking to people?