A new guide details a structured approach for performing large-scale code refactors using AI coding agents like Cursor and Claude Code. The author outlines a four-step process: discovering when to refactor, planning the scope, executing changes, and verifying results. This method leverages LLMs to handle the tedious aspects of code cleanup, making complex refactors more manageable.

The article argues that the need for refactoring has increased with higher coding output from AI assistants, but the effort required has decreased. Key triggers for a refactor include detecting antipatterns like duplicate code or poor separation of concerns, or when implementation speed slows. Using AI models for this task is particularly effective due to their strong working memory.

The core methodology involves using tools like Gemini for architectural discussions and plan mode in Cursor to generate detailed refactoring blueprints. Execution then leans on models like Claude with liberal command whitelisting. This conversational approach with AI agents, rather than simple prompts, is presented as crucial for successful, large-scale code hygiene.