michcia's notes

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

For the last maybe half a year I've been mulling over how I feel about the rise of AI. If you asked me a year ago I would be a complete skeptic and making fun of how great it is going, but I don't feel so simple anymore. I feel like there is a lot to unpack, and I wouldn't be surprised if even in the course of writing all this my opinions change. Since the year started, I've witnessed myself shifting between utter disgust at what the industry is pushing and deep interest in cracking the puzzle of an AI agent that would Actually Be Useful.

So I'll split it up into multiple pieces, starting from the one that felt the most straightforward to write.

I know what online bubbles I gathered around myself, so my main ask is just: if you disagree, vehemently, radically, start by telling me about it. Like I said, my mind is not set in stone, and you can always disown me later.

The image generating models are capable of making good looking graphics, for a number of purposes. The problem though is that it misses the point.

In most cases I'm seeing, the generated graphics serve the role of a glorified clipart image. It's not adding much value, just makes things look "nicer".

Sometimes you can immediately spot the typical AI style, sometimes it's less obvious. But I don't want to be on the hunt for the signs of AI imagery. It helps that most of my exposure to it happens on-corp or, heavens forfend, prefixed with "I asked AI tool name to generate an image of thing and this is what I got". OK but what did you want to convey? If anything? Is it supposed to distract me from you having nothing to say? What is going on here?

It's the same when I see AI used for one-off memes either. Maybe that's cause I still see people making thoughtful commentary or clever jokes with a tool that just gives you three fixed-position text fields on top of an image of your choice, and you can spend 15 minutes adjusting newlines and spaces to get your text exactly where it needs to fall. I think it's great craftsmanship and I appreciate that so much more. There is so much more intentionality in an edit of an ancient meme template covering some letters in the original subtitle with ▮▮▮ unicodes to make funny wordplay, than in a generic looking comic strip that says exactly what you wanted to say, but which the machine laid out by itself with not much human input.

I actually had one meme idea I tried to draft with a genAI tool: the city of Warsaw is running a lottery for people filing income tax here this year, and their domain is placpitwwarszawie.pl which translates to "pay PIT in Warsaw" but the "płać" got un-diacritic-ed into "plac" which to me invokes the image of "the PIT Square/Plaza in Warsaw". I have no idea what the PIT Square in Warsaw could look like. Unfortunately the AI didn't either, gave me a generic looking place with a vaguely correct street sign and a mashup of Warsaw's Old Town views (don't remember exactly, something like the gate of the Uni's main campus). I thought about photoshopping the name myself onto an existing photo of some city square, but I don't see how that would be more interesting, so I just posted a text version of the pun and that was it. I don't think that's a terrible end though.

So this is a case I don't really see myself compromising on.

I plan to keep staying away from generating imagery and music, except maybe to demonstrate to my family (or whoever else needs it) how easy it's become to produce misinformation. I will try to develop my atrophied drawing and instrument playing skills again.

supplementary materials: