pi-autoresearch: Karpathy의 "AI 자율 실험" 아이디어를 범용화한 오픈소스

pi-autoresearch brings autonomous optimization from research labs into developer workflows by generalizing Karpathy's autoresearch concept to any measurable metric—training loss, bundle size, test speed, or Lighthouse scores. The open-source tool addresses production concerns the original proof-of-concept left unsolved: statistical confidence scoring to distinguish genuine improvements from noise, persistent session state, and automated integrity checks to prevent optimization regressions. This positions autonomous experimentation as a practical workflow where developers set objectives and AI agents iteratively improve code overnight, suggesting new possibilities for AI-driven development optimization.