LLMs slot into this conversation in a really interesting way.

The things the author set up are technologically mature enough that, as long as you have the media, or as long as you can get your friends to use it, the self-hosted versions are largely better than the commercial ones. The last decade or so of innovation has really been about figuring out how to monetize these technologies, at the expense of UX.

This is in contrast to LLMs, where the commercial ones kind of wipe the floor with the self-hosted options.

On the other hand, LLMs essentially give average people superpowers for self-hosting mature technologies. My wife used Claude Code to vibe-code an educational game for our five-year-old, tailored to his preferences and the skills he needs to work on (she's a UX designer and now, a couple weeks in, reads enough Javascript to understand when Claude is doing something stupid).

If we want to buy a computer to use a server, write, and host a bespoke family to-do-list / photo store / knowledge base / calendar that syncs my wife's Google Calendar with my .org files ... we are so much more able to do that than we were even two years ago.

Totally agree! I relied a lot on Claude Code to help me with issues I was having setting up hardware encoding in Immich. But I also think – as accessible as LLMs are as a tool – they still don't make the complexities of self-hosting trivial. Maybe they'll get there, but I feel like LLMs really are just another coding abstraction. It's like a programming language. You still need to know how the language works and how the underlying "bare metal" works to make anything reliably.

But, I could be wrong! Thanks for reading and commenting!