Is Learning to Build Software Still Worth It with AI? (My Honest Take)

This post contributes to the understanding of agentic coding harnesses by addressing the practical challenge of rate limiting when interacting with powerful LLMs. It introduces the '2x test' as a framework for evaluating the long-term value of software development skills in the context of rapidly advancing AI. The author also elaborates on the importance of providing AI agents with rich, multi-modal references (like 3D models) and the concept of a 'meat proxy' to caution against uncritical adoption of AI-generated code. The post frames the ongoing relevance of software engineering not in terms of AI's ability to replicate existing functionality (reverse engineering), but in the human capacity for intent, creativity, and building novel systems.