Weeknotes no.1 – atproto, homelab, and AI

i loved the idea of reoccurring weeknotes from other blogs like mijndertstuij.nl.

disclaimer: do not expect this to be overly regular. if there's nothing to note (or no time) i won't post :)

atproto & the fediverse ⁂

i've spent some significant time reading the detailed specs and concepts around the at:// protocol and honestly things are starting to make so much sense. i also stumbled upon tangled.org which is a git-type project that builds the social features on top of at proto – still in alpha stage but very promising. set up one repo over there that is being synced back to github.

if you're also a nerd, go play with pdsls.dev, this has helped me so much in understanding the structure and concepts around personal data servers (PDS) and the atproto protocol.

ai, homelab, hermes 🧠

any time spent on social media bombards me with posts and updates of people building out their hermes setup and since i skipped the whole openclaw hype, i caved and set it up on my homelab mini pc. i made the mistake and first tried to connect it to my own matrix/ tuwunel instance, which in true matrix fashion was quite complicated (don't do that)... telegram on the other hand is super easy and way speedier.

hermes definitely is a nerd and tech tool for now. this is not something you could let a normal person use, the UX just isn't there for it. but it's a great playground to experiment where the future will lead us. and for that it works amazingly well. all of the world's models available in a telegram chat, linked to personal context, an own obsidian vault, session history, skills, mcps, and the whole shebang: an ai nerd's dream ;)

so far i used it to debug some annoying home assistant automations, wrote several documents as a test, and added a few agent cronjobs to regularly check up on things.

links and cool stuff

• artificial analysis has very cool (and detailed!) model comparison charts
• npmx.dev is such a better version of the sluggish npmjs.com website
• mdn now has an mcp server
• top read: Modern Engineering Values
  • fav quote: The best engineers exhibit strong ownership and domain expertise in the area they work in. Agents amplify ownership. Someone who deeply understands a product can now execute faster, while someone without context can generate much more noise. As a result, coordination becomes more expensive. I believe the most effective teams will be small, often two to three people, with clear ownership boundaries and isolated repositories
  • primary learning: context needs to be in repos, not locked in notion or confluence for better agent access
• Ryo Lu (design at cursor) gave a really interesting presentation at their conference about the future of coding and development – i recommend watching it!
  • "AI is collapsing the distance between ideas and reality."
  • "If AI makes execution cheaper but judgment weaker, we do not get a renaissance. We get slop with working buttons."
• slightly concerning, but interesting article about end-to-end encryption on whatsapp
• very non tech: over here it's been up to 38+ºC; a little plate on the balcony with fresh water has been an absolute hit with the local birds 🐥
• i updated my /social page with some recent additions

quote of the week

heard this in a tiktok (sigh) this week:

"If you write a problem down clearly and specifically, you have solved half of it." (Kidlin's law)