The Rise of AI Loops: A New Era in Automated Coding

Boris Cherny, creator of Claude Code, recently spoke at Meta's @Scale conference, where he was asked about the growing trend of AI loops. Cherny confirmed that loops are indeed a game-changer, stating that they represent a significant step forward in the development of artificial intelligence. According to Cherny, the transition from handwritten code to agent-written code has been remarkable, and the introduction of loops takes it to the next level. In a loop, one agent prompts another to write code, creating a continuous process that can lead to significant improvements in code architecture and efficiency. Cherny's own work involves using loops to optimize code, with one agent identifying areas for improvement and another unifying duplicated abstractions. The potential of AI loops is vast, but it also requires a great deal of trust in the technology. As models become more advanced, the use of loops could revolutionize the way AI handles complex tasks. While the concept of recursive loops is not new, the application of non-deterministic logic and the use of sub-agents to control the loop is a departure from traditional computing. The benefits of AI loops are clear, but the costs could be significant, with the potential for rapid token burn and the need for careful oversight to avoid drift and other issues. Despite these challenges, the potential rewards of AI loops make them an exciting development in the world of artificial intelligence.