Boris Cherny, creator of Claude Code, declared at Anthropic's developer conference that he no longer writes prompts. He writes loops. Addy Osmani, formerly of Google, defined it plainly: loop engineering means replacing yourself as the person who prompts the agent and designing the system that does it instead. Peter Steinberger, creator of OpenClaw, preached the same shift publicly. Three high-profile practitioners converged on the same idea within weeks of each other.
The original piece traces this back to software engineer Geoffrey Huntley, who built what became known as 'Ralph loops' roughly a year ago, before the approach went viral in December. By May, major AI coding harnesses shipped the /goal command, letting a single prompt trigger a full loop. Real-world adoption looks like cron jobs, event triggers, auto-opened PRs for newly recorded app issues, and agents that prep notes when an oncall engineer joins an outage Slack channel. But the piece also documents real pushback: agents drifting off-task, human-in-the-loop setups outperforming autonomous ones, and API token costs that make looping expensive fast at companies paying market rates.
The article is worth reading in full because the skeptics are as interesting as the believers. Distinguished engineer Max Kanat-Alexander argues that looping may have been a temporary workaround while tooling caught up, not a permanent paradigm shift. The piece also opens a harder question it does not fully resolve: whether context engineering, not loop engineering, is the more durable skill for developers to build.
[READ ORIGINAL →]