Video yükleniyor...
Video Yüklenemedi
Claude Code creator: "100% of our pull requests at Anrtopic are run by Claude Code. 80–90% of code review too. The feature I’m using the most today is /loops. I’m not prompting Claude anymore - I’m building loops" in 1-hour interview, Boris reveals his setup, which helps him build... show more
999,949 görüntüleme • 4 ay önce •via X (Twitter)
45 Yorum

100% of prs and 80-90% of code review at anthropic is the kind of number that should make people stop scrolling for a second

Yeah, I wish I’d started building with Claude earlier, but the best time is now.

Yeah, if you have imfinite resources, loops is the way, For mere max subscribers, loops is a luxury we can’t afford.

@0xMovez that’s wild, makes me wonder how much time we're wasting prompting when loops could automate so much. gotta try it out myself.

yeah, sooo true bro !

@0xMovez haha right? glad I'm not the only one who thinks so!

Lol, we saw exactly the quality of output - both in code and how the program works itself. Claude code success is 100% in idea(s), not execution. its quality as a software is complete disgrace, so I wouldn't brag about how brilliantly they slopify it

loops is so underrated, glad more people are seeing this

yeah, so true bro ! loops is a real alpha. btw are you using it ?

just starting to experiment with it, still figuring out the right use cases

amazing watch for today ! thanks for share movez

thanks for an amazing article bro !

Wow I hadn’t seen this before , thanks for sharing

you shared amazing article about AMD bro ! Real chad, will read it now and watch the stage

thanks for the support , let’s keep working my friend 🫶🏻

Build loops, not prompts

nice sharing!

thank you Leo ! Btw are you building loops yoursel ?

yeah, ofc👍🏽

real CHAD !

@grok 给我视频yotube 地址

The shift from prompting to building loops is where Claude Code actually gets interesting. Most people are still stuck in the back and forth when the whole point is to let it run.

we really went from writing code to just supervising loops 😅

loop is ass, just do the task and move onto the next, this is just shilling for 10x token usage for no good reason

the scary version of this stat is the engineer who hasn't actually read a diff in months. if 90% is AI, the whole game is whether you still get the 10% that matters. built for exactly that 👇

Boris Cherny, as always, gave out alpha

Another nerd snipe opportunity I see

maybe Claude Code wouldn’t suck so bad if they actually wrote code 😅

funny timing, i literally shipped a loop for this last week. drives ralph to hunt architecture debt and open issues, won't re-propose stuff you already decided on. the stopping logic is the whole game tbh

vibe-coding is officially out, loop engineering is the new elite buzzword

Yeah? And how much money should I burn in /loops? Well, no thanks.

Impressive efficiency using Claude Code for pull requests.

Why people think they are using the same model as us, they had fable for the last 6 months and mythos as well

@grok what podcast was that interview with boris?

Movez, you bring alpha every day

Loops are a real productivity leap—but self-review still needs independent evidence. Contract tests, UI/API automation and CI quality gates keep agent speed from becoming silent risk. My take:

i've seen the future of coding and it feels like winning at my own casino loops aren't the feature, they're the entire table

loop 只是把 prompt 藏进了流程里。该想的一个没少。

the loops thing is what stuck with me too. once i stopped writing one-off prompts and let it catch its own mistakes in a loop everything changed. curious what boris actually puts in his

building systems beats writing prompts every time

building loops instead of just prompting is exactly where the space is heading. curious how they handle ambient authority when 100% of PRs are run by agents. anyone seeing good patterns for restricting blast radius in code review loops?

That’s why we beat ralph in their own game Listen to boris!

And who cares? Loops are a generic code construct. What is so special here?

When you have unlimited tokens to burn, YES, of course you only use loops.

Industry selling more costly workflows, what a surprise
