Optimizing a 200,000-line Python codebase overnight with Gemini 4 Argon- Google Developers Blog

This post details the use of Gemini 4 Argon to optimize a 200,000-line Python codebase (Agent Development Kit - ADK) by identifying and fixing cross-file performance bottlenecks. It describes the process of steering the AI agent using `.agents/skills` rulefiles and `/goal` mode, leading to a 40% reduction in in-memory session prep time and a 21% speedup in cold package imports. Two optimizations were shipped: filtering in-memory session history before copying and loading heavy packages only when needed. Four other proposed optimizations were rejected during code review due to reliance on fragile state or private library internals. The post also outlines the benchmark results and the division of labor between the AI agent and human maintainers.