Operationstactic

Migrate agent prompts from non-English to English for 30-40% token savings

Problem solved: Non-English-speaking teams burn unnecessary tokens because LLM tokenizers are optimised for English vocabulary.

TierCCatalog — Pattern outlined, candidates for deeper work

EffortS · Days

What

Operators benchmarked: rewriting CLAUDE.md, skills, and recurring prompts in English produced ~30-40% token reduction at equal task output — "language-agnostic", a tokenization effect, not a model-quality effect. Pairs naturally with caveman-style verbosity caps.

← All casesC-080