Learnings from 100K lines of Rust with AI (2025)

Editor's take

A developer documented their experience integrating AI, specifically models like Claude and Codex, into a large-scale Rust codebase exceeding 100,000 lines. This exploration highlights the practical challenges and benefits of using AI for code generation, refactoring, and debugging in complex, production-grade software.

The significance lies in moving beyond theoretical AI applications to tangible, real-world software engineering workflows. While current AI code assistants like GitHub Copilot and Amazon CodeWhisperer have gained traction, this detailed account provides a candid look at their effectiveness in a substantial, performance-sensitive project, offering insights for developers and tool creators grappling with the nuances of AI-assisted coding.

Future observations should focus on the long-term impact of AI on Rust's development velocity and code quality. Specifically, whether AI-generated code reduces maintenance burdens or introduces subtle, hard-to-detect bugs will be crucial. The evolution of AI models' understanding of Rust's ownership and borrowing system, a notoriously difficult concept, will also be a key indicator of progress.