Kiro

AI coding assistants often break codebases during refactoring operations like renaming functions or moving files because they treat structural edits as text replacements. Kiro solves this by integrating directly with VSCode's language server protocol and workspace edit APIs, using the same battle-tested refactoring infrastructure developers use manually (F2 rename, drag-and-drop file moves). This approach provides semantic understanding of code structure, automatically updates all references and imports across the workspace, and works language-agnostically through existing language servers. The result is precise, reliable refactoring that maintains codebase integrity without the error-prone find-replace loops typical of LLM-based approaches.

Refactoring made right: how program analysis makes AI agents safe and reliable

The problem: when agents work harder, not smarter