Anthropic engineer: "Stop prompting Claude. Build a system that PROMPTS ITSELF." Here's HOW

You're not supposed to prompt Claude. You're supposed to build a system that prompts itself. That's something I heard from an Anthropic engineer. And I think that is very interesting and I thought I'd share with you guys right now. Okay, like this is something that the head of Claude code has been saying all summer. You know, stop prompting these agents, start building a system that allows these agents to iterate itself. Now, this concept has a name and it's called loop engineering. The whole concept is, instead of you typing a prompt, waiting, checking the results, fixing it, typing another prompt, waiting, that whole process basically itself runs by itself, literally while you sleep. Now, it's interesting as well, the company Stripe actually uses this exact method in their code. If you hit the link in my bio, I've got the whole setup prompt so you can set this up for yourself. Just head over to my Discord. It's in there. Okay? Essentially, here's how that mistake into you engineer. The first mistake is not having a memory file. A memory file, essentially, from the loop starts from zero. So what you want is a file which basically the agent will then record exactly what worked, what failed, what's done, etc. So every single time it runs, this thing gets smarter and it has all the context. The next mistake, there's no split between the maker and the checker. So essentially the maker, the agent that actually writes the code, also checks the code. If there is no split between them, if these aren't separate agents, the issue is if the maker, the agent making the code, also checks the code, it often tends to praise itself. So it's just gonna be biased and it's not going to give you an accurate loop. It's just gonna run into the same errors, it's just gonna accumulate these errors. So make sure the agent that code, the agent that is checking the code, two completely separate agents. No. And mistake number three, there is no stop condition. Okay? Essentially, the loop will just keep running and running and running and running, getting built to oblivion. Okay? So the loop has a condition that actually obviously finished defines when all checks have been passed, this is complete. Head to the link in my bio, the Discord, grab the resource. Let me know how it goes for you. But helps game.