PLC we are committed to improving. as to why we didn't attempt the other two, I think the short answer is that we needed to be realistic and focused in our ambitions for atproto 1.0. you can only bite off so much, and a lot of our core novel ideas are still unproven

multi-writer repos with CRDTs: we talked a bit about that very early on, and local-first stuff is starting to mature. but this would have been way too hard to bite off.

it might be possible to make it more generic/unbundled, and still usable for devs. having modular protocol components others can reuse is definitely a value and aspiration for us.

non-PDS OAuth: there is a real cost to optionality/generality. AT OAuth is already complex and hard to implement or understand, pretty consistently the top app dev complaint about AT, and this would make that hole deeper.

another more urgent thing (IMO) which @mythik.co.uk@bsky.brid.gy has also written about, is FedCM. we'd like to experiment and get more involved with FedCM next year (eg, potentially get involved with W3C process)

this is one of the more thoughtful and balanced things i've read about LLMs in a while! it hits a lot of my feels about social dynamics and use within a team. 576 - Using LLMs at Oxide / RF...

this comparison has probably been made by others, but it just struck me how similar LibsOfTikTok is to the Degenerate Art / Entartete Kunst exhibition

if you haven't seen it, check out the enriched turbostream / sqlite bulk stuff that graze.social provides

thinking about design patterns and data stores for indexing billions of AT records Big Indexing

this is the first time i'd submitted a formula to homebrew and the process was pretty smooth and well documented