Boris Cherny: We Cut 80% of Claude Code’s Prompt

Modern AI development requires an empirical, scientific approach that prioritizes "unhobbling" models—removing restrictive system prompts to allow them to leverage latent capabilities. Boris Cherny, creator of Claude Code, emphasizes that developers should abandon rigid, over-specified instructions in favor of high-level goals and robust verification mechanisms. By treating models as organic entities rather than static systems, builders can unlock significant product overhang, enabling agents to perform complex, long-running tasks like rewriting entire code bases or automating system maintenance. This shift necessitates a move away from traditional software engineering patterns toward dynamic workflows where models orchestrate thousands of agents to verify their own work. Ultimately, the most effective builders in this era are those who experiment iteratively, test at the bleeding edge, and allow models to solve problems that previously required months of human labor.