2/
It's called a red team. Concept from cybersecurity: a team whose entire job is to attack your own defenses before a real adversary does.

Your readers are the real adversary. Not hostile. Just busy. And busy people don't reread paragraph 4 to find your actual point.

3/
Step 1: Create a SEPARATE AI project called "Red Team."

This matters. Your writing AI is your collaborator. Your red team is opposing counsel. They never share a context window.

You don't want your editor and your adversary having lunch together.

4/
Step 2: Load a system prompt that tells AI to attack, not improve.

Five targets: weak arguments, buried ledes, voice drift, dead weight, missing receipts.

One rule: no encouragement. No "strong draft overall." Just the damage report.

5/

What my red team caught in week one:

🧉 Thesis buried in paragraph 4 (twice)
🧉 Same phrase in 3 consecutive posts
🧉 One section that was just the previous section in a fake mustache

None of this showed up when I self-edited.

6/
The fix is three passes, 15 minutes:

Pass 1: Structure (move the lede, cut dead weight)

Pass 2: Arguments (add YOUR evidence, not AI-generated filler)

Pass 3: Voice (read out loud, no AI, would-you-say-this-to-a-friend test)

7/
The order matters.

Structure → Arguments → Voice

If you polish voice before structure, you'll fall in love with sentences that need to be cut.

Build the house before you pick the paint.

8/
After three weeks, something shifts. You start catching problems while writing. The damage reports get shorter. Not lower standards. Better instincts.

Full system with exact prompts:

Part 1: articles.cowritewithai.com/p/you-need-an-enemy

Part 2: articles.cowritewithai.com/p/now-put-it-back-together https://mastodon.social/@nickquick/116352838878164797