Boris Cherny (Claude Code's creator) said something that stopped me cold: "I don't prompt Claude anymore. I write loops and the loops do the work. My job is to write loops." everyone clipped the line. almost nobody built what he described. so i watched all 4 minutes and rebuilt... his exact his exact setup claude code + loops + a dynamic setup that runs itself. worth more than any $10,000 vibe-coding course. watch the talk. then read this. everything you need to know about loops to actually run it ↓ bookmark both. this is your weekend.show more

is loop engineering the same as "/loop"

pretty much and of course cretaing /loop workflows when to use and not to use

@Av1dlive loops doing the work? That's wild. Feels like we're just scratching the surface with AI. I need to rethink how I'm using Claude now.

Loops doing the work is exactly the shift that feels wild at first. Once the verify step starts catching its own mistakes the whole thing compounds differently. The article lays out the full stack that makes it sustainable.

You've moved from prompting Claude to architecting with Claude. That's not a productivity gain, that's an identity change.

@TheAIShrink Can you explain more

amazing work bro

Thanks noisy

clipped the quote and went back to writing prompts like nothing happened fr

Clipping the quote and moving on is the real test of whether the article landed. Most people just read and forget the loop structure. The verify step is what actually sticks when you come back to it later.

the verify step is where most loops quietly die tbh you build the loop then skip verification because it feels like overhead

not sure the gap is really the full-4-minutes vs the clip might be wrong but most products just don't have obvious loop candidates to start from. did what you built end up looking like his version or did the pattern need totally rethinking for your context?

Yes I am using it right now

everyone is clipping it but nobody is actually building

Yes that’s so true

you think this is his best interview? or which video do you think is the most useful?

The Karpathy interview is strong but the one on agentic workflows might be more directly useful for what you're asking. It gets into the spec writing and verification layers that actually move the needle. The article builds on that exact idea.

The “loops” framing is underrated. Once you stop treating the model as a one-shot answer engine, the hard parts move into the runtime: state, permissions, retries, verification, and knowing when to hand control back to the user. For browser/web tasks, the harness often matters as much as the model.

Nice share

私はこれに関して何も言いたくない。ただ驚きでいっぱいです。

You're one of the few people who actually studies the source Quality research as always 🔥

@QuilCraftedWord Thanks Kira anything particular you want me to write on

nice sharing!

thanks leopard

Ai agent slack when done right becomes a super power

That’s so true

W sharing

thanks a lot please watch the entire clip

When you say loops, are we talking about simple retry logic or something more akin to a state machine driving the agent's interaction? I've found basic retry loops often fall short for complex tasks.