I tried using the caveman speech skill with Claude AI after seeing recommendations on Reddit, and it truly made a difference in token usage. The technique involves updating the system to use terse, short sentences—like a caveman would talk—reducing unnecessary words. For instance, instead of a normal request that consumes around 180 tokens, the caveman style drops it to roughly 45, saving up to 75% of tokens. This method uses fewer and simpler tokens per command, cutting down the cost significantly when performing repeated web search tasks or direct instructions. The installation was straightforward, and once configured, the AI responds with blunt, clear replies while still fulfilling requests effectively. From personal experience, the key is to embrace the minimalistic caveman language style—three to six words per sentence with no fillers. It took a little practice but improved performance and efficiency substantially. For users who rely heavily on token-based AI platforms, incorporating this skill can lead to cost savings and faster interactions. Overall, this caveman mode is a clever, practical hack to optimize AI communication, especially useful when you want to manage your token budget smartly. I recommend commenting "caveman" to get the skill and trying it out yourself to experience the benefits firsthand.