"Why waste time say lot word when few word do trick." That's Kevin Malone from The Office, and it's now doing prompt-engineering work inside at least one developer's CLAUDE.md file - with measurable results.
CLAUDE.md is a configuration file that Claudee Code](/tools/claude-code/) reads before every session. It lets you set persistent instructions for tone, behavior, and output format without repeating yourself every time. Adding a single Kevin quote under a "Response and Writing Guidance" section turned out to reduce response verbosity, tighten agent-generated documents, and cut context usage - the amount of text Claude can hold in memory at once. Once that budget fills up, older parts of the conversation get dropped, so verbose responses eat through it faster than necessary.
The interesting part isn't just that it works - it's why. Explicit instructions like "be concise" or "keep responses short" tend to get overridden when Claude senses the topic is complex enough to "deserve" a longer answer. A cultural reference sets a tone instead of issuing a command. Kevin's grammar is broken, but his meaning is unambiguous. Claude apparently reads it as a style anchor rather than a one-time directive.
This is a variation on the "caveman prompting" approach that has been circulating among Claude Code users - using stripped-down language as a persistent style guide to fight the AI's default tendency toward over-explanation. The Kevin version adds personality, which may explain why the resulting outputs feel natural rather than just curt.
If you use Claude Code and find your sessions full of five-paragraph explanations when you asked for a bullet point, this is a 30-second test: add the quote to your CLAUDE.md under a writing guidance header and see what your next session produces. The context savings alone make it worth trying.