Understanding Loops · A Series

The habit AI builders switched to, explained for everyone else

Most people use AI the slowest way there is: type a request, wait, fix, ask again — all by hand. The people who build AI stopped. This series explains what they do instead — specs and loops — in five short chapters, written for people who don't code.

"I do not prompt Claude anymore. I have loops running that prompt Claude and figure out what to do. My job is to write loops."

Boris Cherny · Claude Code, Anthropic — public explainer: 17.7M views
Understanding Loops · the series
Five short chapters, built to be read in order. Start at 01 and you'll have the whole picture.
01 From wishing to guaranteed resultsWhy the same AI returns slop for one prompt and exactly what you wanted for the next — the four decisions that fix it, with worked examples. 02 Prompts ask. Loops finish.The flagship: how AI finishes work without you — the mechanism, the failure data, a run you can watch, and the exact file. 03 Do you need a loop?Honest answer: often, no. Three questions decide it in ten seconds. 04 Twelve practices that hold upThe reference page — named patterns for specs and loops, each with a copyable artifact. 05 Paste. Pick a mode. Copy.How the tool does all of the above for you — the whole flow, animated.
New here? Start at 01 — the chapters build in order.
Cheat Codes
A separate series: standalone guides that crack open one tool and show you exactly how to run it. No order — take the one you need.
01 Unlocking ClaudeHow Claude Code actually works — two workers, one loop, and no memory — plus the three controls that hand you the wheel.