What's the best programming language for coding agents?
Summary
The article analyzes claims about token efficiency differences between dynamic and static programming languages for coding agents using language models. It reviews multiple evals (e.g., Zstd, Pandoc) and argues that token efficiency is not universally tied to language type and depends on task size, test design, and context. It emphasizes methodological caveats, the influence of language popularity, and the need for broad, careful evaluation before drawing strong conclusions.