Speaking to AI Agents like Cavemen Saves 65% of Tokens. We Test.
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
JetBrains ran a rigorous A/B benchmark of the 'Caveman' token-compression skill on Claude Code using SkillsBench (86 tasks, ~240 trials, ~$106 spend). The skill claims 65% output token savings, but testing shows only ~8.5% savings on real agentic work — because code, diffs, and tool calls dominate agent output and Caveman correctly leaves them untouched. The good news: no measurable quality degradation was found across 82 paired tasks (p=0.82). Cost savings track the token reduction (~10%) but are fragile due to single-trial variance. The 65% figure applies to chat-style Q&A, not coding agents. Verdict: safe to use, fun, but don't expect significant savings on daily agentic tasks.
