Yes!! I was just complaining about this: bsky.app/profile/nate...
I'm sure things are being sand bagged on recent models. opus-4-5 is 10x faster in some cases now over opus-4-6
Posts by Nate Butler
I guess it was unclear from the original screenshot. Other model is opus-4-5
> ✻ Churned for 5m 15s
4-6 ended up taking 10x the amount of time 😅
Results were basically the same
Same task, same prompt... why does opus-4-6 take 3,4,5,6x longer to do things??
Alfred never went anywhere 🙂
Title slide: One Developer, Two Dozen Agents, Zero Alignment: Why we Need Collaborative AI Engineering
The meme of two astronauts looking at earth. The one behind has a gun pointed at the other's head. First says "wait, alignment is the bottleneck?". Second says "always has been"
Putting together the final touches on slides for my talk at AI Engineer conf this Thursday @aidotengineer.bsky.social
Talking about how we need realtime, multiplayer, collaborative agentic dev tools. Not isolated, single player terminal instances.
And showing one we've built at @githubnext.com!
Audiobook is stellar too!
This just wasn't possible until like, yesterday
right now everything in the world is telling you to go faster, ship more, add that feature, start another project
so i'm actively working on feeling ok not doing any of that
Unfortunately cache read/write killed it. Editing anything in history kills cash, driving up costs a lot.
Anthropic nudged us away from the raw text approach due to that.
One of the original concepts of the first agent in @zed.dev that we never got to was the agent rewriting the history like this as you went - trimming out side quests and making the back scroll more of a canonical document
Sometimes I miss the whole conversation just being an editable text document.
Not to suggest the problem as the fix but… we’ve been exploring fixes for these problems at @githubnext.com - Don built the repo assist workflow to help with these exact types of problems: dsyme.net/2026/02/25/r...
It could also be tweaked to close those ai generated PR suggestions as well 🙂
Nice investigation!
maybe the antis were right all along 🙄
tfw you just barely catch in the scroll claude getting irritated it can't solve a problem with some test seeds, so it just decides to encode into the test to skip them...
utterly amazing
Zeta2 is here. 30% better acceptance rate than Zeta1. 200x more training data, LSP-powered context, faster predictions, open weights. Try it now in Zed.
We didn't just improve the model. We rebuilt the entire data pipeline behind it: zed.dev/blog/zeta2
Somehow people got conned into thinking that is what good design is 😅
Been getting back to shipping some components for gpuikit - maybe a new release will come sometime this month...
lmao, even claude uses yolomode these days
I assume most of open source heads this way tbh. Which is a little sad, but understandable at this point. Lots of great contributions to @zed.dev came about through open prs.
4-6 feels like 4-5 got lobotomized - and people look at me crazy when I say it is bad
We're hiring an Open Source Engineer! 🦀
Build features, review PRs, shape how contributors engage with Zed.
Interesting problems, Rust codebase, small team, fully remote.
Your work ships in days and users notice.
Know someone? Are you someone? 👀
zed.dev/jobs/open-so...
Have the dubious honor of receiving my first open claw recruitment email 😅
time to start thinking about better filtering tools for email...
100%. No notes!
🧑🍳
design like this could just be... normal again.
more like this please 🔥
every.to/source-code/...
making a GUI IDE in 2026 is an increasingly meditative act