Developers are making AI talk like a caveman to cut token costs

Developers have started using a tool that turns AI assistants from chatty conversationalists into caveman-like responders that communicate in short, stripped-down sentences.

According to 404 Media, instead of replies like "You're right to push back, I was wrong," the model produces something closer to "Hulk smash"—just code, commands, and facts, without the conversational filler. Companies are adopting the tool for a simple reason: reducing AI costs by cutting the number of tokens their models generate.

The tool, called Caveman, was created by developer Julius Brusse in April. Technically, it is a small configuration file in the SKILL.md format that can be added to Claude Code, Codex, Gemini, and dozens of other AI agents with a single command.

Caveman strips greetings, filler words, and introductory phrases from AI responses while preserving everything that actually matters, including code, commands, links, numbers, and technical terminology. As Brusse describes it, the model stops behaving like a polite chatbot and starts acting like a concise tool.