Boris Cherny, creator of Claude Code, recently discussed the significance of AI loops at Meta's @Scale conference. Cherny believes loops are a crucial step in AI development, allowing agents to prompt other agents to write code. This approach enables continuous improvement of code architecture and identification of duplicated abstractions.
Cherny's statement highlights the potential of AI loops in handling real work, with models getting better rapidly. However, this shift also requires significant trust in AI and can be expensive due to high compute requirements.
The concept of AI loops is not entirely new, as recursive loops have been used in computer science for a long time. Nevertheless, the application of AI loops in agentic AI is a notable development, with potential benefits outweighing the costs for certain problems.
