Benzi – A harness reading LESS source code to write better quality code
SMRTR summary
Benzi is an AI coding agent that parses entire repositories into a precise, queryable symbol map using tree-sitter before answering any question. Instead of dumping code into a context window, it resolves call edges, data flow, and class hierarchies upfront. On SWE-bench Verified, it resolved 78.2% of 500 real bugs at roughly $0.095 each — reading far less source code than competing tools like Claude Code or Cursor.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article