Caveman to Lessen Cognitive Strain

With agentic coding, my work as a software engineer has changed drastically. Before LLMs, my job involved a heavy dose of writing code; now I operate at a higher level, steering the process instead. It's like being a CPO/CTO with everhappy and energized code monkeys.

As a result, I read and write far more specs than before, or, well, have LLMs do it for me. Either way, it's a big shift toward non-code work. Since I work at an American company, English is my daily working language. It's also the default language of software, so I use it even in my own projects.

The problem is, English isn't my first language. Even though I'm fluent, it's still more mentally taxing than my native Finnish. As AI speeds up development, the volume of English I consume grows quickly. My ability to consume the foreign language becomes a bottleneck.

Caveman to the Rescue

Like many of you, I, too, have experimented with different Claude Code skills. A few weeks ago, a new one: Caveman by Julius Brussee, made the headlines.

Despite its tongue-in-cheek style, I've found Caveman genuinely useful. It compresses LLM replies into fewer words, using a blunt, caveman-like tone (hence the name). Here's an example from the repo:

Normal Claude:

"The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle. When you pass an inline object as a prop, React's shallow comparison sees it as a different object every time, which triggers a re-render. I'd recommend using useMemo to memoize the object."

Caveman Claude:

"New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."

This is not only easier on the brain, but it also saves tokens (and thus money)¹. Less filler means faster iteration. If I need more detail, I can always prompt more.

Despite working mainly in English, I've started using Finnish with agents (though I instruct them to write code and docs in English). Officially, Caveman supports only English and Chinese, but I've found work real well with Finnish — it makes it short, direct, and to the point.

Taking it Further

Caveman offers three compression levels:

1. Lite — minimal compression, great for concise professional text.
2. Full — default, drops articles and sounds "grunt-like".
3. Ultra — maximum compression, heavily abbreviated.

I've been using the Full level in English and Ultra level in Finnish and have been very happy with it. Reading less to get the same information makes everything feel clearer. I don't feel as mentally fried as soon into a vibe-coding session as before.

There's also the /caveman:compress skill, which compresses files. While standard Caveman makes Claude write less, /caveman:compress makes it read less. This can further save you those valuable tokens.