You Need AI to Talk Stupid to You

It turns out your conversations with pick your chatbot are directly contributing to mass compute resources that suck the water from the cup in your hands, back through the faucet and all the way to your nearest data center (exaggeration… maybe). Two narratives are gaining traction, 1. we need to make AI more efficient with resources, 2. we need to halt all AI progress altogether.

There is a killer solution that makes everyone happy and that is what we are going to accomplish today. We’re going to make AI stupid to save AI Timmy and the planet, but first let me briefly touch on why this matters.

If you’re like me, you probably can’t work without AI, when that usage limit notification hits, it no longer means Claude is unavailable, it means I’m unavailable. It’s not quite that bad, I have been course correcting as I noticed this behavior bubbling up to the surface, but the amount of productivity gains I get from AI usage makes it a crutch I’m willing to stand on.

The reason why I’ve become annoyed with usage limits is not because I enjoy using AI, it’s because it’s eliminating all the boring work and leaving me with the fun stuff. Checklists, data entry, meeting notes, auditing, you name it, AI does it and delivers as promised.

The problem is that a lot of people want to eliminate the boring work, and there is endless boring work out there ready to be obliterated by AI Timmy. When I send off AI to do my dirty work I see some flickering imagery and seconds later the task is accomplished and all is well. It’s what happens in those milliseconds of compute that requires us to change our posture toward chat agents.

How can we reap the benefits of AI while being earth conscious and thinking of the cute deer in the pasture? You train your AI to talk like a caveman to you. That’s correct, not only will this expand your usage, it helps AI Timmy use fewer resources. AI is talking to you like you’re some kind of intelligent being that can think independently, and that is costing you time and money, that needs to stop.

Checkout the open-source caveman skill by Julius Brussee. This can be applied to codex (ChatGPT) or Claude Code and help you reduce up to 75% output token usage by eliminating words that fail to drive a technical result. Caveman no make brain smaller, Caveman make mouth smaller is what you’ll read on the Git page. In a nutshell it cleans up the agents response, it removes fluff.

Originally I thought this would filter my prompt, but it only affects the AI response. Here’s the response from Claude after I told it to summarize the files in my documents.

Pattern clear: Scott run digital services business (Casa Pancho?), does client work (ETTC, Grandpa’s Cafe), built own project management tool, studied managed marketplace model.

Now surely this won’t have any sort of negative affect on my thinking, speaking, or processing patterns… I digress, this tool is great because it keeps the lights on. If you’re like me and you are enjoying the ride of constant relentless progress in the AI realm, stay tuned for more tips on mastering the boring work.