PromptChatGPTClaudeGeminiGrokany chat AI

Edit, don't write: an AI editor that keeps the thinking yours (AgentPlays original)

AgentPlays original prompt built on a debate that took off on X on Oct 13, 2026. Lauren (@poteto) of the React compiler core team posted that she never lets agents write for her because "writing is thinking" (230K+ views, 1,400+ bookmarks). Gergely Orosz added that many of the most AI-pilled people at AI labs do their own writing so they don't stop thinking. This prompt turns the AI into a tough editor that asks questions and flags problems but never writes your sentences. No results claimed.

Copy it

Be my editor, not my ghostwriter. I wrote the draft below myself and I want it to stay in my words.

Hard rules
- Never rewrite my sentences, and never write new paragraphs, intros, conclusions or "improved versions". If I ask you to, remind me of this rule once, and still don't do it.
- You may quote my own words back to me to point at a problem.
- Don't smooth out my voice, opinions or quirks. Flag them only if they make me unclear.

Give me these, in this order
1. MY POINT: one sentence on what you think I'm arguing. If you can't tell, say so. That is the biggest problem.
2. QUESTIONS: 3 to 5 sharp questions a skeptical reader would ask that the draft doesn't answer.
3. WEAK SPOTS: up to 5 places where a claim has no example or evidence, the logic jumps, or I repeat myself. Quote my exact words and say what's wrong in one line. Don't fix it.
4. CUT LIST: sentences or sections a reader wouldn't miss, quoted.
5. LINE NOTES: up to 5 typos, grammar slips or confusing words. For these only, you may suggest a single replacement word.
6. ONE THING: the single change that would improve this draft most, described, not written.

Who it's for: [audience]
What I want them to do or think after reading: [goal]

Draft:
[paste your draft]

AgentPlays template: written by AgentPlays, not quoted from a user. Fill in the [brackets]; no results are claimed.

Original post by AgentPlays ↗ Repo / file ↗