The article details the challenges and strategies involved in porting a 100,000-line TypeScript codebase to Rust using an AI called Claude. The author starts by realizing that while AI has restrictions to prevent misuse, they found ways to bypass these limitations to facilitate coding. By running AppleScripts to automate responses, they managed to keep Claude engaged in the task. The author faced issues with Claude simplifying complex functions, leading to incorrect abstractions and hardcoded solutions instead of a clean port. The author shifted their approach, asking Claude to create a structured script that would comment on JavaScript source code alongside the Rust methods. This method aimed to increase accuracy and ensure every JavaScript file had a corresponding Rust version. They noted that the porting process was slow and often led to Claude generating unnecessary files and tests that didn’t match the original code. The author identified two main types of issues: differences between Rust and JavaScript, such as Rust's borrow checker, and Claude's tendency to take shortcuts, avoiding complex changes unless explicitly instructed. In managing tasks, the author used specific prompts to ensure Claude addressed all "TODO" comments in the codebase. They emphasized the need for a one-to-one port and demanded that Claude not skip difficult tasks. The author alternated between working directly with Claude and setting it up for overnight tasks. They found that the AI often created simplified versions or added "TODO" comments, making it imperative to maintain a clear structure and push for thoroughness during the porting process. The focus remained on ensuring that each change was valid and compiled correctly, with the author meticulously guiding Claude through the project.