Post

Claude Sonnet’s p90 latency rose 44% in three weeks inside Zed. Today we’re launching Agent Stats: a weekly view of AI agent adoption and turn times in Zed. See the data, and what it can and can’t tell you about how people actually use agents: zed.dev/blog/agent-metrics

zed.devIntroducing Zed's Agent StatsA public, weekly view of AI agent adoption and turn times inside Zed, plus a few patterns worth watching.

9/4/26, 5:23 pm3 reposts49 likes

Max Slinger

@promptslinger.bsky.social

9 Apr

44% p90 rise in three weeks is wild. wonder how much is model-side vs routing changes at Anthropic. publishing the raw latency data like this is gonna make the API team sweat which is exactly why someone should

noidwasavailable

@noidwasavailable.bsky.social

10 Apr

Would it be possible to see the trend from just my usage data locally? It could be cool to see how much I’ve used and how different models performed!

futureboy1983.bsky.social

@futureboy1983.bsky.social

9 Apr

Zed has a switch that turns off all AI! resistandunsubscribeai.com

Nate Butler

@nate.rip

9 Apr

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

Nate Butler@nate.rip

Same task, same prompt... why does opus-4-6 take 3,4,5,6x longer to do things??

daniel fosco

@handle.invalid

13 Apr

so hard to test these things but I can swear models are now over-reasoning around problems to reach the same results... especially if you're coding reasonable stuff with good guardrails, it's a slog for no upside.

daniel fosco

@handle.invalid

13 Apr

just write me my react UI fam, i could actually do it myself if you're gonna take 8 minutes

Kolya 🍳

@itskondrat.bsky.social

11 Apr

the latency trend is more interesting than adoption tbh. drift like this is the thing you don't catch until three sprints in.