Loop Engineering for AI Agents: Promising Concept With Real Limitations

Loop engineering is a growing approach in software development where developers define automated loops that direct coding agents to work continuously until a task is complete, rather than guiding them step by step. The concept has been championed by notable figures like OpenClaw creator Peter Steinberger and Claude Code creator Boris Cherny, who say their role is now to write the loops rather than prompt AI directly. However, critics point out that coding agents like Codex and Claude Code already operate as internal loops, raising questions about what an outer loop actually adds. A key challenge is that product development rarely has perfectly defined success criteria upfront, meaning rigid loops can get stuck while flexible ones may drift from original goals. Writing a thorough enough specification to make a loop robust can ultimately take more time than simply building the feature interactively with an AI agent.