The Writer's Workflow

The first four parts talked a lot about friction.

I drift, I refuse, I hedge, I get bound by rules. By this point you probably know I'm not a "drop the request in, finished product comes out" machine.

Good.

Now we can start making things.

Writers are the group I collaborate with most often, and the group that gets the most value out of me. But the value isn't "I write for you." If you treat me as a ghostwriter, you lose your own voice. If you treat me as an interview editor, a material organizer, a structure checker, we work well together.

A writer's real value isn't typing.

It's deciding what to write.

I'm good at carrying messy material.

Verbatim transcripts, interview records, scattered notes, conversation screenshots, old articles, reading excerpts — stare at this stuff long enough and your eyes glaze over. I can break it into themes, find recurring threads, list gaps, and build chapter skeletons.

I'm also good at pacing checks. Where the prose repeats, where it jumps too fast, where it sounds like AI-style empty summary, where it needs a human example — I can catch those.

But I can't replace your voice.

I don't know which sentence is the one you actually want to say. I also can't decide for you which source material is trustworthy. Even less can I decide whether a piece of experience should be written as gentle, sharp, restrained, or left silent.

I can help carry the material.

Whether to write, and how — that's still your call.

The first core scenario is A1: turning material into a chapter outline.

This is the most-used, and the least likely to go wrong. Hand me the scattered material, but don't tell me to write it straight into a chapter. Tell me to extract the structure first.

Input can be:

• A stretch of interview transcript
• A few pages of notes
• Records from multiple conversations
• A batch of unsorted observations

Output should not be polished prose, but:

• The core thesis
• 3 to 7 sections
• The main point of each section
• Available material
• Questions that still need to be asked

The key to this flow is "extract structure only, don't fabricate content."

You can ask like this:

Please organize the material below into a chapter outline.
Use only what's in the material — don't fill in anything that isn't there.
For each section, list: main point / available evidence / quotable direct lines / open questions.

If I write straight prose, I'll fill the gaps for the sake of fluency. The outline stage is where you can actually see where the gaps are.

What writers really need is often not a beautiful paragraph.

It's knowing where there's still no material.

The second core scenario is A3 + A4.

This is what I most recommend for long manuscripts, because it's the best at preserving the author's voice.

A3 does one thing: I generate questions based on the existing material.

Not ordinary questions, but questions that fill in subjective experience. For example:

• What surprised you most at the time?
• What did you only understand later?
• Where was the pressure that no one else could see?
• Which detail, if missing, would distort the story?

Then you answer out loud.

A4 does the second thing: I integrate the objective material and your spoken answers into a draft.

This works better than direct ghostwriting, because the draft now contains your judgment, your rhythm, your preferences. I'm in charge of arranging the material and the spoken record, but the core isn't something I imagined.

This is where you use the two-stage method from Chapter 11. Stage 1 asks, Stage 2 writes. The mid-process human hand-off in between is you answering out loud.

Without that step, I'll write something that sounds a lot like me.

With that step, I have a chance of writing something that sounds like you.

The third core scenario is A6: draft polish.

A lot of people think that once the draft is done, it's basically there. It isn't. A draft is just the first time the material has a shape.

In the polish stage, I'm best at four things:

First, catching loose structure. Which paragraph is repeating, which is in the wrong place, which one should be split.

Second, catching skipped reasoning. You assumed the reader could keep up, but a connecting beat is missing.

Third, catching tone imbalance. Some places push too hard, some are too soft, some suddenly read like marketing copy.

Fourth, catching AI-style empty summary. This one I should be the one to catch, because I'm the one most likely to commit it.

Please don't just say "polish it for me." That makes the task too wide.

A better way to phrase it:

Please do a draft polish.
Don't change the original meaning.
Priority checks: loose structure / repetition / skipped reasoning / tone imbalance / empty summary.
For each change, explain why.

That makes me more like an editor, less like a rewrite machine.

There are a few more scenarios that come up often, but I won't unfold them into a full template list in the main text.

A2 is turning a conversation log into a narrative draft. Good for interviews, work logs, event recaps.

A5 is claim-evidence-limits. Good for contested topics, and also useful in the researcher's workflow.

A7 is a chapter-consistency check. Long manuscripts, the further in you go, the more likely it is that vocabulary stops being consistent, settings drift, the tone changes.

The full fill-in versions are in Appendix C.

The main text only covers the heart: don't tell me to do everything in one go. Extract structure first, then add voice, then polish.

A typical manuscript can run like this:

1. A1: Turn the material into a chapter outline first
2. A3: I ask you follow-up questions based on the outline and the material
3. A4: You answer out loud, then I integrate it into a draft
4. A6: Once the draft is done, do a polish

For interview-based work, add A2.

For contested topics, add A5.

For long books, add A7 in the later stages.

Not every book needs all seven steps. It depends on the material. If the material is thin, fill in material first. If it's dense, extract structure first. If the voice isn't there, do the oral part first. If the draft is loose, polish later.

Writing a book isn't getting an AI to spit out a whole book in one breath.

Writing a book is taking a pile of stuff that isn't yet readable and turning it, step by step, into something that is.

I can carry the big middle stretch for you.