Cut the fluff. Get to the point.

AI assistants love pleasantries. 'Sure! I'd be happy to help you with that. Let me take a look at the code and see what I can find.' That's twenty words of nothing before the actual answer starts. Multiply that by every response in a long session and you're wasting a lot of space on manners instead of substance.

It strips out the filler. No articles, no pleasantries, no hedging. Responses become fragments: 'Bug in the login form. Fix: check the validation step.' Instead of 'I found an interesting issue in the login form component. Let me walk you through what I discovered and propose a solution.'

Important things stay normal. Code, commit messages, and security notes don't get compressed. Only the conversational filler goes away.

AI tools have a memory limit. Everything in your conversation takes up space in that memory. When the memory fills up, the AI starts forgetting what you discussed earlier. Caveman mode saves that space for actual information instead of burning it on 'I'd be happy to help' sixty times in a session.

By hour two of a long session, the space savings are significant. Early context stays in memory because the AI didn't waste room on manners.

Memory is finite. Spend it on information, not politeness.

Set up a rule that activates at the start of every session. The rule is simple: drop filler words, use fragments, get to the point. Say 'normal' or 'stop caveman' when you want the polite version back.