Member-only story
Loop Engineering Is Trending.
Claude Code Shipped It Months Ago.
Most of what’s going viral about “loops” right now describes infrastructure Claude Code already built for you.
“Loop engineering” is having a moment on my timeline this week the idea that you shouldn’t be prompting agents turn by turn, you should be designing a loop that runs them for you.
It’s a good idea.
It’s also, if you’re using Claude Code, mostly already done.
I wrote a while back about the orchestration half of advanced Claude Code ultracode, dynamic workflows, fanning a hard problem out across a fleet of subagents that work in parallel and cross-check each other.
That solves the wide problem: a task too big for one context window.
There’s a second half of “advanced” that solves a different problem the long one: a task that takes more turns than you want to sit there approving, one at a time, for hours.
Claude Code has two built-in commands for exactly that, and almost nobody outside the people who read changelogs has touched either of them.
