Meet "loop engineering": The next evolution in AI coding isn't a better prompt, it's a system that prompts itself | TechSpot

The field of AI development is experiencing a significant shift from simple prompting to a more sophisticated approach known as "loop engineering." This evolution is primarily driven by the escalating costs associated with continuously running complex AI agent systems, which rapidly consume tokens. Developers are therefore moving towards designing more efficient, self-sustaining setups where AI systems can operate with minimal ongoing human input once initiated. Loop engineering, as described by experts like OpenAI engineer Peter Steinberger, involves setting up a framework and a clear goal, allowing the AI agent to work autonomously through a recurring workflow, rather than requiring step-by-step human prompts. Tools such as Claude Code and OpenAI's Codex already utilize this concept with features like the "/goal" command, enabling agents to persist on a task until completion. The benefits extend to cost efficiency, as detailed by Google Cloud director Addy Osmani, who cautions that while sub-agents burn more tokens, strategic implementation of loops can be invaluable. This shift means the focus is less on crafting individual prompts and more on structuring tasks, managing agent communication, and enabling unsupervised, repetitive processes, effectively treating AI agents like employees assigned specific roles and responsibilities.