Building a Self-Improving AI Code Factory

This article details the author's experience building a personal AI code factory using Claude, o3, and Sonnet AI models. The core principle is "fix inputs, not outputs": instead of directly patching generated code, the author adjusts plans, prompts, or agent combinations. The factory iteratively improves through planning (o3), execution (Sonnet), and verification (o3 and Sonnet), using Git worktrees for parallel development. The author shares scaling strategies, such as creating specialized agents and enforcing consistent code style. The ultimate goal is an AI system that autonomously generates, verifies, and improves its own code.