Stop Writing Prompts, Start Designing Loops — Running Four Claude Code Boxes 24/7

The video opens on what looks like a flashy demo but is creator Benjamín Cordero's daily reality. Four physical machines, each running Claude Code nonstop, and he hasn't typed a prompt in days. He walks past them while they keep working.

The roles are split cleanly. Box 1 continuously scrapes his community's posts and reindexes them into a rack, retraining on which posts performed and which didn't. Box 2 goes out once an hour to hunt trending angles and videos on YouTube, dropping them in his morning queue with possible approaches. Box 3 monitors the world and the community at large. A fourth ties it all into a running to-do list, and yet another is in the middle of a multi-hour build right now. Each box has its own subscription and runs Claude Code 24/7. He didn't build this for the video — it's been his workflow for months, longer than most people have been talking about this approach.

The lead on Claude Code inside Anthropic said it on stage: he no longer writes prompts, he writes loops. The video walks the pattern through five concrete pieces. The main loop is build → fix → verify → goal check, repeating until the exit condition fires. The fifth piece is memory — a 'brain' that retains what falls out of the active context window so the loop doesn't forget itself.

The implementation is straight Claude Code slash skills (Codex has a loop function too, and the same pattern works there with the right prompting). `/goal` handles bounded loops — those with a definite finish line. `/loop` handles open-ended ones. The author observes that when Anthropic ships new features, they mostly land in this area — Dream and the rest are refinements on how loops work — which suggests the platform is moving toward this pattern as the primary mode.

The closing thesis is crisp: stop operating tools. Orchestrate agents that operate tools. Then orchestrate the agent orchestration with loops. Move one level up the abstraction stack. Stop prompting, start designing.

A prompt is spent in one message. A loop keeps running while you're not at the keyboard. So you stop being an operator and become an architect. This isn't a technique, the author says — it's a mentality shift. If you can write and you can build, the programming language of the future is natural language. He's been pushing that thesis for over a year and a half on the channel; loops are where it finally cashes out into a concrete workflow. Channel mantra to close: don't buy, build, and build better with loops.