Context Master Blog

Semantic vs Heuristic Code Graphs for AI Agents

This article compares three different approaches to building context for coding agents: the baseline grep/ripgrep approach most agents still use today, semantic graphs (LSP- or compiler-based) and heuristics-based parser graphs (tree-sitter). We evaluate which of the approaches produces the best context for Coding Agents like Claude, Cursor, Codex or Github Copilot.