Carmack proposes AI-optimized coding styles for weaker models

This is a fascinating shift in perspective, moving from optimizing code for human readability to optimizing for AI comprehension. It introduces the concept of "AI-friendly" code architecture and linting rules. It highlights a potential meta-use case where frontier models are used to identify best practices that empower smaller, cheaper models. It raises interesting questions about whether code optimized for transformers would still be intuitive and readable for human developers, and suggests that standardizing coding styles specifically for AI could significantly lower the barrier to entry for automated code generation and maintenance.