i had this realization a few weeks ago and its spot on. recently when getting hungry, i want to go work out. unfortunately too hot to run in austin without waking up super early or on the trail in the dark (no thank you), but as it starts getting cooler i'm really excited
57
cry me a river and go back to your suburb of Ohio
I was surprised how bad it was just walking from the hotel to the Rails World venue. One strung-out guy wanted to fight me for some reason, another was swinging wildly and swearing profusely. Lots of zombie vibes too. Whatever Austin is doing, it's not working.
78
hey @shashankgoyal95, qwen/qwen3-reranker-8b on OR is throwing 429s pretty consistently from fireworks. do you think yall could put more effort into yall's embeddings and reranking offerings?
2
56
claude.ai has a sitemap called "artifacts-touch-grass-sitemap.xml"
claude.ai/sitemaps/artifacts…
71
hey @charliermarsh @astral_sh, can i get your opinion on our usage of uv for our big monorepo and how it could be improved?
this is our layout:
python/packages/* — 40 packages
python/apps/{api,agent,mcp} — 3 apps
frontend/* — typescript apps consuming a version-controlled openapi spec generated from the api
we use nx (@nxlv/python) for the dependency graph, so lint/build/test only run for affected projects, and the graph crosses languages.
when the openapi spec changes, the frontend apps are marked affected too. affectedness is attributed per-project via per-project lockfiles, so each of the 43 python projects has its own uv.lock, and dependency versions drift across them
we haven't found a way to use uv workspaces here. one root uv.lock means every lock change marks all 43 projects affected, which defeats nx. is there a supported way to share resolution/constraints across independent lockfiles?
is this/could this be the fault of our nx plugin? this is the largest codebase i've ever worked on so i'm not super familiar if we're doing this "the best way". open to alternatives, too
111
Carter retweeted
better coding and spend less!
can't promise you will stop yelling at the model but this is a significant step forward
Coding agents with memory made up to 65% fewer corrections at up to 52% lower cost.
Knowledge Pages. Self-updating docs that synthesize your project's decisions straight from memory, so your wiki heals itself instead of going stale.
In depth breakdown:
3
6
420
Carter retweeted
Introducing celld: a self-hosted, distributed Durable Objects and Workers implementation
- celld = V8 + S3 + SQLite + LTX+ Tokio
- programmed with exactly the Cloudflare Workers/DO JavaScript APIs and config
- order of magnitude cheaper at scale
- writes are durable before they're acknowledged, RPO=0
- only depends on an S3 compatible bucket - no other infrastructure
- ~4 MB memory per cell instance overhead
- hibernated cells cost ~nothing, just their S3 storage
- open source, Apache 2 licensed
120
325
2,989
325,724
Spotify's new AI app, Studio, isn't built on the Claude Agent SDK. or anything from OpenAI.
It's built on @pidotdev, an MIT-licensed agent framework by @earendilworks. 🧵
1
4
234
