What is loop engineering, AI's next big trend?
What's the story
As artificial intelligence (AI) systems become more sophisticated, a new trend is emerging in the tech industry: loop engineering. This concept involves creating automated loops that can control AI systems, assign tasks, and manage workflows with minimal human intervention. The idea is to build a system that handles repetitive interactions with AI so humans don't have to step in at every turn.
Evolution
Loop systems take the reins from prompt engineering
Loop engineering is a major departure from the current trend of prompt engineering, which has been the norm since ChatGPT popularized conversational AI in late 2022. Prompt engineering involves carefully crafting requests to get the best results from large language models. However, some software developers are now exploring ways to automate this process altogether with loop systems that can guide and supervise AI agents until a job is done.
Advocacy
Tech leaders champion loop engineering's potential
Leading figures in the tech industry are advocating for this new approach. Boris Cherny, co-founder of Anthropic and head of its Claude Code team, recently revealed that he no longer manually writes prompts but uses an AI-driven coordination system instead. Peter Steinberger, creator of OpenClaw, has also encouraged developers to focus on designing systems that can manage AI agents automatically rather than spending too much time crafting prompts.
System elements
Components of a loop system
Google Cloud director Addy Osmani, a major proponent of loop engineering, explained that such systems are made up of different components. Automations let tasks run continuously instead of once. Worktrees allow multiple agents to work simultaneously without interfering with each other. Skills provide instructions and project knowledge while plugins and connectors give AI access to external tools. Memory helps retain information between sessions by storing project histories externally. Sub-agents introduce checks and balances by assigning different responsibilities to separate models.
Use cases
Developers explore loop engineering's practical applications and challenges
Developers are already testing practical applications of loop engineering. Steinberger, now working at OpenAI, has created a Codex loop that automatically wakes up, manages repositories, and distributes work across threads. Osmani warns that running several agents continuously can consume large numbers of tokens and significantly increase costs, advising developers to deploy sub-agents only when a second opinion justifies the expense.