Boris Cherny, creator of Claude Code, told Anthropic's developer conference he no longer writes prompts. He writes loops. Within weeks, Peter Steinberger of OpenClaw and former Googler Addy Osmani said the same thing publicly. Loop engineering, defined by Osmani as 'designing the system that prompts the agent instead of prompting it yourself,' went from niche practice to trend fast enough that The Pragmatic Engineer went looking for what it actually means in production.
The piece traces a specific origin: software engineer Geoffrey Huntley's 'Ralph Wiggum' loop, shared a year ago, which went viral in December and seeded the terminology. By May, major AI coding harnesses shipped the /goal command, letting a single prompt run a full loop. In practice, most developers are using loops as glorified cron jobs and event triggers, not transformative new workflows. Some are abandoning them entirely, citing agent drift, better results with a human in the loop, and token costs that compound fast at API pricing rates. Distinguished engineer Max Kanat-Alexander argues the loop was a temporary hack while tooling caught up, not a permanent paradigm shift.
The most valuable part of this article is not the definition of loop engineering. It is the gap between the conference-stage rhetoric and what developers report actually works. The original includes specific loop patterns that do hold up, including auto-opening PRs for newly logged app issues and prepping context for oncall engineers joining outage channels. It also introduces whether context engineering, not loop engineering, is the more durable skill. Read it for the friction, not the framing.
[READ ORIGINAL →]