Return to blogging

The last time I sat at the keyboard — or more precisely, used it as a tool to write blog content — was in a completely different reality. AI language models were not yet ubiquitous, and the world, not just the tech world, simply looked different without them.

However, my return to blogging has nothing to do with life-altering dramas, a midlife crisis, or other topics that would fit tabloid pages. That said, I must admit that the authors of sensational stories about “hiding in the closet at night” or other life catastrophes after the arrival of AI were probably one of the happiest social groups. They could finally — or rather had to — make decisions about radically changing their lives.

Then again, who knows? Perhaps the absurdity of those publications has reached such a level that no model, however advanced, could generate them. I have no idea.

Although the world has changed significantly since then, my reason for returning to blogging is, as always, the desire to test certain solutions in practice. The Gatsby framework that powered the previous version of the site has been replaced by Astro. Most of the code was generated by GPT-5.2, Claude, or Gemini models, while Blender, used for creating graphics, evolved in the meantime from version 3.x to version 5 and beyond.

These are substantial changes. To elaborate on the mentioned models and explain the rationale behind these technological leaps, it’s enough to point out that none of them, on their own, managed to handle what I consider a very simple task. I started with Copilot, which failed across the board. In this context, GitHub’s decision to stop aggressively pushing paid subscriptions doesn’t surprise me at all. Copilot remains on my setup, but only in the free version and solely as a VSC plugin.

I also tested Cursor, Windsurf, and a few other options. In the end, the latter remained in the running. I have to admit, however, that implementing this small project in Astro cost me a lot of time and persistence. It may sound trivial, but try finding a ready-made solution online that natively combines i18n, tags, pagination, and a smooth search function. It turns out that this combination is practically unheard of — at least I haven’t come across one.

It’s certainly not a technological barrier, but apparently either the documentation is rather dense, or the demand for such a feature set is exceptionally rare.

In any case, the blog is coming back to life. While it’s hard for me to say how long it will last, one thing is certain: the topic of hated language models and the solutions built on them will, like it or not, make up a significant part of it.