Loop Engineering: Why Programmers Should Design Systems That Prompt AI, Not Do It Manually

In mid-June 2026, Claude Code creator Boris Cherny sparked a major shift in AI development thinking when he stated at Acquired Unplugged that he no longer prompts AI agents manually — instead, automated loops handle that task for him. Days later, OpenClaw founder Peter Steinberger amplified the idea on X, writing that developers should design loops to prompt their agents rather than doing it themselves, a post that garnered 8 million views in a single day. Google engineer and O'Reilly author Addy Osmani subsequently formalized the concept as 'Loop Engineering' in an O'Reilly Radar article, defining it as replacing yourself as the person who prompts the agent. A Loop Engineering system follows a four-step cycle — Act, Observe, Reason, and Repeat — allowing automated programs to assign tasks to AI agents, review outputs, and decide whether to continue or stop, even while the developer is away. Steinberger noted that over 40% of his code is already written by such loops, and warned that 90% of companies he works with make the mistake of not using verification loops at all.