Two weeks ago our team of six engineers switched entirely to caveman mode for all Claude Code sessions. Here is what happened.
Week 1: Adjustment
The first day felt weird. Responses seemed curt. After two days, nobody noticed anymore. The directness became the baseline.
Token Usage
Our weekly output tokens dropped from ~4.2M to ~1.1M. That is a 73.8% reduction. Monthly cost dropped significantly. Small team, big multiplier at enterprise scale.
Code Quality
Zero change. We compared outputs on 30 matched tasks. Technically identical. In two cases reviewers actually preferred the caveman answer — the directness made the root cause clearer.
Recommendation
Switch your whole team. Adjustment period is two days maximum. Savings are permanent. Caveman not complicated. Just efficient.