Finding the way through a 100,000-line codebase with AST — How to understand large codebases with Kiro Code Intelligence | Amazon Web Services

AWS argues that text-only AI coding tools fall short on legacy, multi-language codebases because they miss structural relationships and context.

The piece highlights AST-based analysis as a better approach for searching, tracing, and refactoring code across large repositories.

Kiro Code Intelligence uses tree-sitter and LSP to build a two-layer system for more reliable code understanding.

The goal is to reduce hallucinations, improve navigation, and make AI assistants genuinely useful on real-world codebases.