CHOW: Chat-Oriented Writing

I used to stare at blank pages. Now I have a different problem: AI will happily fill those pages for me, but my brain stays empty. CHOW - Chat-Oriented Writing - is my answer: using AI to think better, not just write faster. The written artifact is a side effect. The real product is clarity in your head. This post covers why writing equals thinking, the risk of outsourcing that thinking to AI, and the four thinking modes - Madman, Architect, Carpenter, Judge - that help you partner with AI instead of just delegating to it.

• CHOW is Thinking
  • Why Writing = Thinking
  • The Risk: Outsourcing Your Thinking
  • For You vs. For Them
• The Four Thinking Modes
  • The Madman: Unleashing Creative Chaos
  • The Architect: Structuring Your Thoughts
  • The Carpenter: Crafting Your Words
  • The Judge: Polishing Your Work
  • Key Principles for AI-Enhanced Thinking
• The Art of the Back-and-Forth
  • Filtering Is Easier Than Creating
  • The Collaboration Patterns
  • AI Slop vs. Human Sludge
• Trampoline Prompts
  • What Makes a Prompt a Trampoline
  • Trampoline vs. Delegation
  • Trampoline Prompt Patterns
• Deep ideas
  • Explainers: AI-Generated Interactive Understanding
  • Generating the content for the reader
  • New monetization strategies

CHOW is Thinking

Why Writing = Thinking

The act of putting ideas into words forces clarity. Vague notions that feel profound in your head reveal themselves as muddled when you try to explain them. This is why writing matters - not for the artifact, but for the thinking it produces.

CHOW builds on the principles of Chat-Oriented Programming (CHOP):

The Risk: Outsourcing Your Thinking

AI makes writing easy. That’s the obvious part. The less obvious part: the thinking that writing forced on you - the struggle to clarify, to find the gap in your logic, to articulate what you actually mean - that was the point.

When AI writes for you, you skip the struggle. And the struggle was doing the work.

This is why CHOW matters. The goal isn’t “get AI to write stuff.” The goal is “use AI to think better.” The difference:

For You vs. For Them

Every piece of writing serves two audiences:

For You (Understanding) - Writing to figure out what you actually think. The reader is future-you or maybe no one. The goal is clarity in your own head.

For Them (Being Understood) - Writing to transfer understanding to someone else. The reader is a real audience. The goal is clarity in their head.

These aren’t separate activities - they’re interleaved. You can’t explain something clearly until you understand it yourself. And the act of trying to explain often reveals you didn’t understand it as well as you thought.

AI can help with both, but you have to know which mode you’re in.

The Four Thinking Modes

The Madman, Architect, Carpenter, and Judge aren’t just writing phases - they’re thinking phases. Each serves two purposes: helping you understand, and helping others understand.

AI can be a powerful partner in each phase. Let’s explore how:

The Madman: Unleashing Creative Chaos

The Madman phase is all about uninhibited brainstorming and idea generation. Here’s how AI can enhance your creative madness:

Prompt Storming

Use AI to generate multiple angles on your topic:

• “Show me 10 different ways to approach this topic”
• “What are some unconventional perspectives on this?”
• “Help me explore the extreme edges of this idea”

Idea Expansion

Let AI help you go deeper into each thought:

• Feed it your raw ideas and ask for related concepts
• Use it to find surprising connections between topics
• Ask it to play “what if” with your concepts

Research Assistant

While you’re in creative flow, AI can:

• Find relevant quotes and references
• Suggest related topics to explore
• Gather supporting evidence for your ideas

Remember: In this phase, quantity beats quality. Let both your mind and the AI run wild.

The Architect: Structuring Your Thoughts

The Architect phase is about organizing your chaos into coherent structure. AI can help by:

Pattern Recognition

Feed it your Madman output and ask:

• “What themes do you see in these ideas?”
• “How might we organize these thoughts?”
• “What’s the most logical flow for these concepts?”

Outline Generation

Work with AI to:

• Create multiple possible outlines
• Test different organizational frameworks
• Identify gaps in your structure

Examples:

• Restructured my retirement guide’s TOC using AI to improve flow and consistency - see the conversation and changes
• Organized book references using consistent conventions - see the changes and conversation

Audience Analysis

Use AI to consider:

• How different readers might approach your content
• What background knowledge they’ll need
• Where you might need more context or explanation

The Carpenter: Crafting Your Words

In the Carpenter phase, you’re building your first draft. AI can be your assistant by:

Draft Collaboration

Generate initial sections based on your outline:

• Generate initial sections based on your outline
• Suggest transitions between ideas
• Offer alternative phrasings

Examples:

• Added a security considerations section to CHOP documentation - see the changes and conversation

Style Consistency

Help maintain your voice throughout:

• Help maintain your voice throughout
• Check for tone consistency
• Suggest improvements while preserving your style

Enhancement Suggestions

Identify areas for improvement:

• Identify places for examples or analogies
• Suggest where to add data or quotes
• Recommend areas that need more detail

The Judge: Polishing Your Work

The Judge phase is about critical review and refinement. AI can help by:

Technical Review

Ensure technical accuracy and polish:

• Check grammar and style
• Identify repetitive phrases
• Ensure consistent terminology

Content Analysis

Verify the quality and flow of content:

• Verify logical flow
• Check for gaps in argumentation
• Ensure all claims are supported

Reader Perspective

Consider the audience’s viewpoint:

• Simulate different reader reactions
• Identify potential points of confusion
• Suggest clarifications where needed

Key Principles for AI-Enhanced Thinking

1. Keep the Phases Separate:
   • Don’t let AI’s capabilities tempt you to mix phases
   • Complete each phase before moving to the next
   • Use different prompts for different phases
2. Maintain Creative Control:
   • AI is your assistant, not your replacement
   • Use AI to enhance your ideas, not generate them entirely
   • Trust your instincts when AI suggestions don’t feel right
3. Iterate with Purpose:
   • Use AI feedback to improve each phase
   • Don’t get stuck in endless refinement
   • Know when to move forward

The Art of the Back-and-Forth

The magic of CHOW isn’t in what AI produces - it’s in how you collaborate. Understanding this changes everything.

Filtering Is Easier Than Creating

Here’s the core insight: your taste is the bottleneck, not your idea generation.

Creating from scratch is hard. Staring at a blank page, pulling ideas from nothing - that’s cognitively expensive. But looking at five options and saying “that one, but tweak it this way”? That’s easy. Your brain is wired for pattern recognition and preference, not ex nihilo creation.

This flips the traditional writing process. Instead of:

1. Think hard → produce draft → revise

You get:

1. Describe what you want → AI proposes options → you filter and refine

The AI handles the generative load. You handle the taste. Both play to their strengths.

The Collaboration Patterns

AI proposes, you filter. Ask for five alternatives, not one answer. “Give me 5 title options” beats “What should the title be?” Options give you material to react to.

Fast loops over perfect proposals. Short exchanges, quick iteration. Say “3” or “mix 2 and 5” and move on. Momentum matters more than getting each step perfect.

Think out loud, let AI structure. Verbalize half-formed ideas. “I want something about… maybe the feeling of… or like when you…” The AI catches these fragments and helps crystallize them.

Refine, don’t just accept. AI suggestions are starting points. Take “For Real” and make it “For This Human.” The magic happens in your refinement.

Systematic sweeps. Go through your piece methodically - every heading, every section. Focused decisions beat scattered attention.

AI Slop vs. Human Sludge

Everyone complains about AI slop - that generic, over-enthusiastic, emoji-laden output that screams “a robot wrote this.”

But have you looked at LinkedIn lately? Twitter? Any social media? Human sludge is everywhere. Corporate jargon. Passive voice. “Learnings” and “deliverables” and “at the end of the day.” Humble-brag posts. Engagement-bait threads. Throat-clearing paragraphs that say nothing. Humans produce absolute garbage without any AI help.

Here’s the thing: AI can be a hell of a lot better than human sludge. At least AI doesn’t have ego getting in the way. It doesn’t pad content to sound important. It doesn’t write to impress rather than communicate.

The bar isn’t “as good as a human.” The bar is “as good as the actual content humans produce” - and that bar is surprisingly low.

But of course, the real value isn’t replacing human sludge with AI slop. It’s using AI to augment: take human sludge and make it good, or take high-quality human content and make it even better. That’s what CHOW is actually about.

Trampoline Prompts

Most prompts land flat. You ask, AI answers, you move on. Your brain never left the ground.

A trampoline prompt is different. It bounces you — the human — into thinking you wouldn’t have done on your own. The AI’s response isn’t the point. Your reaction to it is. You land somewhere higher than where you started.

This is the core mechanic of CHOW working well. When the prompt is a trampoline, the conversation becomes a thinking tool. When it’s not, you’re just generating text.

What Makes a Prompt a Trampoline

A trampoline prompt has three properties:

1. It creates a gap. The prompt surfaces a tension, contradiction, or question you hadn’t noticed. Your brain has to work to close the gap.
2. It requires your knowledge. The AI can’t answer it alone — it needs what’s in your head. This forces you to articulate things you’ve been carrying around as vague intuitions.
3. The answer surprises you. Not the AI’s answer — yours. When you respond to a trampoline prompt, you discover you think something you didn’t know you thought.

Compare: “Write me a summary of this meeting” is a delegation prompt. Your brain idles. But “What decision did we actually make in this meeting, and does everyone agree?” — that’s a trampoline. You have to think. You might realize you don’t know the answer.

Trampoline vs. Delegation

Both are useful. But they do different things.

Delegation prompts have their place — sometimes you genuinely need AI to handle the generative load (see Filtering Is Easier Than Creating). But if every prompt you write is delegation, you’re outsourcing your thinking. CHOW dies.

The ratio matters. If I look back at a session and every prompt was delegation, I didn’t learn anything. If at least a third were trampolines, I probably came out sharper than I went in.

Trampoline Prompt Patterns

Here are patterns I use, especially with Larry:

The Mirror — Force yourself to see your own pattern.

• “I’ve said I’d do X three times now. What’s actually stopping me?”
• “What would someone who reads my last five journal entries conclude about my priorities?”

The Contradiction — Surface the tension between what you say and what you do.

• “I say I value Y, but my calendar shows Z. What’s going on?”
• “Where am I fooling myself right now?”

The Pre-mortem — Think backward from failure.

• “It’s six months from now and this project failed. What happened?”
• “What’s the most likely way I’ll sabotage this?”

The Outsider — Borrow a perspective you don’t naturally take.

• “How would my manager describe what I’m struggling with?”
• “If I were coaching someone in my exact situation, what would I tell them?”

The Zoom — Change the altitude of your thinking.

• “Forget the details — what’s the actual question I’m trying to answer?”
• “I’ve been thinking about this abstractly. What’s the concrete next action?”

The Gap — Find what’s missing from your reasoning.

• “What am I not considering?”
• “What question should I be asking that I haven’t asked?”

The best trampoline prompts feel slightly uncomfortable. That’s the bounce.

Deep ideas

Explainers: AI-Generated Interactive Understanding

When AI writes code or explains a concept, the default output is text. But text is passive — you read it, nod, and forget. Explainers are interactive visualizations that let you do the thing instead of just reading about it. And AI can build them for you in minutes.

Simon Willison calls this interactive explanations — one of his Agentic Engineering Patterns. His example: he had an AI build a word cloud tool in Rust, but the algorithm description — “Archimedean spiral placement with per-word random angular offset” — meant nothing to him. So he asked the AI to build an animated HTML page where you can step through the algorithm frame by frame, watching each word try positions in a spiral until it finds one that fits. That made it click.

This is a weapon against cognitive debt — the understanding gap that grows when AI writes code faster than humans can comprehend it. The same AI that creates the debt can pay it down: ask it to build you an interactive explanation of what it just built.

Consider the difference:

• Static text: “Religions evolved from shared roots, with Buddhism emerging from Hinduism around 500 BCE”
• Explainer: An explorer where you click through traditions, see connections, and discover relationships yourself
• Static text: “ChatGPT launched in November 2022”
• Explainer: A timeline visualization showing exactly how long ago each AI milestone happened, updating in real-time

Text gives you information. Explainers give you understanding. See the full explainers page for great examples from Nicky Case, Red Blob Games, and others — plus the AI-powered workflow for building them.

Generating the content for the reader

The author now provides the raw content, and it can create content on demand optimized for the reader, which can include pivots like:

• Should it be interactive
• What age level to write for
• What interests does the read have.
• What format (text, tweet, video, audio)

Aerial talks about this nicely here

For computer folks this is like write path vs read path in data systems

New monetization strategies

Instead of selling static assets (book, video) - you provide the content and it is generated for the reader.

See also: AI FAQ - Open questions: Am I outsourcing my thinking? Is AI actually thinking? Who decides what AI should say?