Everyone demos Claude Code the same way plan, review, accept The engineer who built it doesn't work like that anymore "I don't prompt Claude I write loops that prompt Claude for me" 5–10 sessions at once sub-agents by the hundreds phone, not laptop You're one level behind the guy... who built the tool Full breakdown belowshow more

@Skaly__Bull wait, did you just say you're running 5-10 sessions at once on a phone? that sounds like chaos but i'm intrigued!

phone not laptop is the real flex here

people really out here paying for the privilege of training their own replacement

writing loops that prompt the model for you instead of sitting in the chat. that’s the real usage shift.

Phone not laptop? How is that even possible

5-10 sessions at once changes what coding even means

If loops and sub-agents become the default, how does that change what Claude needs to get right? Do you end up optimizing for orchestration over raw capability, or do you need both to scale?

Claude Code is becoming an operating system

That mental shift is probably more important than learning another prompt trick.

Writing loops that prompt Claude instead of prompting it directly, that's the real shift.

That loop pattern is the real shift. Once you stop hand-holding every prompt, the bottleneck becomes your own reading speed.

seeing the workflow shift like that makes the whole coding mental model flip

So this is what they're doing with AI today, impressive

What does the loop actually look like — task queue, retry logic, context windows per sub-agent? That's the part nobody demos.

in one day i'm became software engineer

bác dùng điện thoại còn em vẫn loay hoay với laptop 😅