The Coming Loop

Armin Ronacher reflects on the rise of 'harness loops' — outer loops that orchestrate AI coding agents beyond a single session — and his deep ambivalence about them. He acknowledges loops work well for code porting, performance exploration, and security scanning, but argues they amplify LLMs' tendency to produce overly defensive, complex, and hard-to-understand code when applied to long-lived production systems. He warns that as attackers, competitors, and reporters increasingly use automated loops, defenders and maintainers will be forced to adopt them too. His core concern is that codebases built and maintained entirely by loops may become incomprehensible organisms that assume machine participation as a permanent maintenance requirement — creating cognitive and operational dependencies that could be catastrophic if access to powerful models is lost. He concludes that the question is no longer whether loops will dominate, but how to retain human judgment, good engineering principles, and meaningful oversight within a looping future.