Skip to content

Instantly share code, notes, and snippets.

@escherize
Created June 22, 2026 17:43
Show Gist options
  • Select an option

  • Save escherize/28a67fb9cb4a2f7fbf9f1f439b9ef8e5 to your computer and use it in GitHub Desktop.

Select an option

Save escherize/28a67fb9cb4a2f7fbf9f1f439b9ef8e5 to your computer and use it in GitHub Desktop.
Why the pyramid-principle skill is worth using

Use the pyramid-principle skill whenever you write executive-facing prose; it pays for itself there and nowhere else.

Situation. You write memos, docs, PR descriptions, reports - things a busy reader must grasp fast.

Complication. LLMs (and most writers) default to context-first: setup, reasoning, conclusion buried last. Reader hits paragraph 4 before the point. "Be concise" and "be MECE" don't fix it - vague advice doesn't change model behavior.

Question. Is a structured skill worth the overhead over just writing?

Answer. Yes for executive writing, no for casual chat or code - three reasons below.

It fixes the LLM's worst writing habit: burying the lead.

The skill forces BLUF - bottom line up front. Answer in the first sentence, support next, data last. Directly counters the context-first default that pushes the conclusion to the bottom where readers never reach it.

Its two-stage split keeps accuracy while leading with the answer.

"Lead with the answer" appears to fight "think before you commit." The skill resolves the conflict: reason bottom-up in a hidden scratchpad, then emit top-down. Accuracy lives in the reasoning (cited tests: reasoning-first ~47% vs answer-first ~33%); pyramid structure lives in the output. Without the split, leading with the answer turns reasoning into post-hoc rationalization.

Its rules are countable, so they actually change behavior.

Vague directives ("be concise," "be MECE") don't move an LLM. The skill swaps them for enumerated, affirmative rules: one bottom-line sentence, three support arguments, named dimensions instead of the word "MECE," positive phrasing over prohibitions (models read negation poorly). Plus two cheap pre-send checks - first-sentence test, read-titles test - that catch a buried lead before shipping.

Skip it for trivial replies, code, and terse exchanges.

The structure earns its overhead only when a reader must extract an argument fast. On a one-line answer or a code edit, the scaffolding is pure cost.


Bottom line: worth it when output is executive-facing writing and the reader needs the argument in 30 seconds; skip it everywhere else.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment