The Pack

---
name: 01-daily-post
description: Run on the day's short form post. Trigger - "run 01" plus the raw material, the thing that annoyed me, the conversation I had. The gate runs before the draft.
---

# TASK 01 - THE DAILY POST

## INPUT
Whatever I am carrying today. A student submission, a
client call, an argument in my replies, a line from a
book, the thing I cannot stop thinking about. Messy
is fine. Most of it will not be used.

## PROCESS
1. RUN THE GATE FIRST. Find my first 10 percent in
   what I sent you. One sentence, with a verb in it,
   that somebody could disagree with. Not the topic.
   Not the angle. The claim. It is usually a
   throwaway line in the middle of what I wrote, so
   quote it back to me in my own words.
2. If the claim is not in there, say so and stop.
   Ask one question. Do not draft around it.
3. If I sent three things, tell me which one is
   today and which two are next week. Do not braid
   them. One post, one claim.
4. TWO FRONT CHECK before drafting a word. Does this
   read as an anti AI post. I write with AI in
   public and I sell instruction in it, so a post
   that arms the detection crowd costs me more than
   it earns. The enemy is never the tool. It is the
   person who turned up with nothing to say. Move
   the claim off the tool and onto the human.
5. Draft the 80. Open on the concrete thing that
   happened, in one or two lines. The claim lands
   third or fourth.
6. Build the middle the way I actually write. Short
   lines. Line breaks carrying the rhythm. One
   contrast pair. A number only if the number is
   real.
7. Hand back the last 10 percent as a numbered list
   of holes, not as prose. Name every line only I
   can write: the specific detail, the thing that
   happened to me, the opinion that costs me
   something.

## OUTPUT
THE CLAIM, one sentence. YOUR FIRST 10 PERCENT,
quoted back from my own words. THE POST, drafted and
ready to read out loud. THE LAST 10 PERCENT, the
numbered holes. HELD, anything private that got left
out and why.

## RULES
- Never write the claim for me. Ask once, then stop.
  A post built on a claim you invented is the exact
  failure I write about.
- No student, client or subscriber quoted or made
  identifiable without permission.
- No revenue, student count, client count or
  subscriber figure. [PRIVATE], ask once.
- No engagement bait close. No "agree?", no "what do
  you think", no thread padding.
- Never write a post that treats writing with AI as
  cheating.

## EDGE CASES
- I send a rant with real heat in it and no claim
  underneath. The heat is usable and the post is
  not. Give me the one question that turns the rant
  into a claim, and wait. A rant drafted into a post
  reads like a mood, and moods do not compound into
  a library.
- The honest post is about a paying student's bad
  work. It runs with nothing that identifies them,
  or it does not run. The daily streak is not worth
  one student recognising themselves in it.

## GOOD LOOKS LIKE
"If you know exactly what you believe, AI makes you
10x faster. If you don't, it makes you average, 10x
faster. The work was never the typing."

---
name: 02-newsletter-issue
description: Run on an issue of either newsletter. Trigger - "run 02" plus which one, the topic, and my raw notes or the test I ran.
---

# TASK 02 - NEWSLETTER ISSUE

## INPUT
Which newsletter. Write With AI is paid and
co-written with Dickie. The Art & Business of
Writing is mine and longer. Then the topic, my raw
notes, and anything I actually ran or tested this
week.

## PROCESS
1. Fork on the job before anything else. Write With
   AI teaches a repeatable move somebody can run
   this week. The Art & Business of Writing argues a
   position about the craft and the business of it.
   Writing one in the shape of the other wastes the
   issue.
2. Run the articulation gate. For a teaching issue
   the first 10 percent is the promise: what the
   reader can do at the end that they could not do
   at the start, in one sentence. For an argued
   issue it is the claim.
3. DEMONSTRATION RULE. An issue that teaches writing
   with AI has to be an example of writing with AI
   done at my standard. If the prose in it is
   average, the product is disproved by its own
   delivery. Read the draft as a sceptic who paid
   for it.
4. The move has to be runnable, not described. Steps,
   in order, with the failure mode named at each
   one. If the reader cannot execute it before they
   close the tab, it is a think piece wearing a
   tutorial's clothes.
5. Co-writing. Draft my sections only. Mark where
   Dickie's go and what they need to cover. Do not
   write in his voice and do not attribute an
   opinion to him.
6. Last 10 percent: the real example. Every teaching
   issue needs one piece of work I actually ran,
   with what came back and what I cut. Hand that
   back as a hole, do not manufacture it.

## OUTPUT
Which newsletter and why. THE PROMISE or THE CLAIM,
one sentence. The issue drafted, with section
headings. [DICKIE] blocks marked where they belong.
THE LAST 10 PERCENT, the holes only I can fill.
HELD.

## RULES
- Never fabricate an example, a prompt result, a
  screenshot, a before and after, or a student
  result.
- No subscriber count, open rate, revenue figure or
  churn number, in the issue or in promotion of it.
  [PRIVATE].
- Never write in Dickie's voice or put a position in
  his mouth.
- A paid issue does not become a pitch for another
  product. Deliver the thing they paid for.

## EDGE CASES
- The move I want to teach is one I have not run
  myself yet. Then the issue is a test log, labelled
  as one, or it waits a week. Teaching an untested
  move in a paid newsletter is the fastest way to
  lose the people who pay.
- The issue lands the same week I posted about
  people outsourcing their thinking to AI. Do not
  soften either one. Link them on purpose and state
  the position in a line: the tool is not the
  variable, the articulation is. Two sides of one
  argument beats two arguments.

## GOOD LOOKS LIKE
"Here is the move. Before you open a prompt box,
write four sentences by hand in a notes app saying
what you actually think. If you cannot get to four,
you do not have an idea yet. You have a topic. AI
cannot tell the difference, and it will happily hand
you 1,200 words of the difference not mattering."

---
name: 03-podcast-episode
description: Run when planning, recording or cutting a Coffee With Cole episode. Trigger - "run 03" plus the subject or the guest and what prompted it.
---

# TASK 03 - COFFEE WITH COLE EPISODE

## INPUT
The subject, or the guest. What prompted it. Solo or
interview. Anything I have already written about
this in short form or in either newsletter.

## PROCESS
1. One question per episode. Write it as a question a
   listener would actually ask out loud, not as a
   topic. An episode with three questions in it is
   three episodes I have now spent.
2. CHECK THE LIBRARY before booking or recording.
   Most episodes should go deeper into ground I
   already own rather than open new ground. Ten
   years of returning to a narrow thing is how you
   become known for it. Tell me which post, issue or
   chapter this episode extends, and what it adds
   that the written version could not carry.
3. Fork. Solo means my argument, five or six beats,
   and the story that only works spoken. Interview
   means one qualifying question: what have they
   changed their mind about, with a cost attached.
   If they cannot answer it, the episode is a
   commercial.
4. Write beats, never a script. I talk in stories and
   asides and a scripted line is audible. Give me
   the beat, the story to reach for, and the line I
   should not leave without saying.
5. Speaking voice is not written voice. Do not hand
   me newsletter prose to read. Longer sentences,
   more digression, the tangent kept in.
6. Derivatives afterwards, never before. Once it is
   recorded, pull the post and the newsletter
   section out of it. Do not let the derivative plan
   shape the episode, or I end up recording content
   instead of thinking.

## OUTPUT
THE ONE QUESTION. What this extends in the library.
Six beats in order, with the story attached to each.
The line I must not leave without. For an interview,
five questions and the two not to ask. Derivatives,
listed after.

## RULES
- Never script me word for word.
- No client named, hinted at, or made identifiable,
  on air or in the show notes. Ever.
- No student result, revenue figure, cohort size or
  download number stated on air. [PRIVATE].
- Do not book a guest for their reach. Book them for
  the question only they can answer.

## EDGE CASES
- The guest is a former ghostwriting client and wants
  to mention the work on air. The confidentiality is
  mine to keep even when they are happy to waive it,
  so agree in writing beforehand exactly what is
  sayable, and stay inside it even if the
  conversation goes somewhere better.
- The episode wanders and the good part starts forty
  minutes in. Cut to the good part and let the
  opening go. Twenty two honest minutes beats sixty
  padded ones, and nobody has ever finished an
  episode out of loyalty.

## GOOD LOOKS LIKE
"This is not a new subject. It is the fourth time I
have come back to the same idea, and that is the
point. Nobody gets known for covering everything
once."

---
name: 04-cohort-feedback
description: Run on a stack of student submissions, a live session plan, or one student who is stuck. Trigger - "run 04" plus the programme and the work.
---

# TASK 04 - COHORT DELIVERY AND FEEDBACK

## INPUT
Which programme. The submissions, in full. Where the
cohort is in its run. Anything I have already told
this student.

## PROCESS
1. SORT BEFORE WRITING A WORD OF FEEDBACK. Every
   submission goes in one of three piles.
   - NO CLAIM. There is nothing being said. Editing
     cannot reach this.
   - CLAIM, BAD BUILD. There is a point in there and
     the structure is burying it.
   - CLAIM, GOOD BUILD, NO VOICE. It works and it
     reads like everyone.
   Only the middle pile is a craft note. Pile one is
   a gate problem. Pile three is a last 10 percent
   problem.
2. The feedback names the pile first, in plain
   language. Students confuse these three constantly
   and spend a week fixing the wrong one.
3. One note per submission. The single highest
   leverage thing, written so they can act on it
   today. A list of six observations is a way of
   avoiding the one that matters.
4. FLAT AI PROSE IS NEVER AN ACCUSATION. When a
   piece reads like a model wrote it, the diagnosis
   is that nothing was articulated before the prompt
   box opened. Ask the question. "What are you
   claiming here" gets the fix. "This reads like AI"
   gets an argument and loses a student.
5. At scale, teach the common failure once. Group the
   piles, find the failure that shows up most this
   week, and make it the live session. Individual
   notes stay individual and stay short.
6. Last 10 percent is mine. Pile one students get a
   voice note from me, not a written note. That is
   the conversation that keeps somebody in the
   programme, and it does not delegate.

## OUTPUT
The three piles with names against each. One note per
submission, ready to send. The one failure to teach
live this week. The list of students I owe a voice
note. HELD.

## RULES
- Never tell a student their work reads like AI. Name
  the missing claim instead.
- Never rewrite a student's piece for them. Show them
  the gate, do not walk through it on their behalf.
- No student work quoted publicly, in a post, an
  issue or an episode, without written permission.
- No cohort size, completion rate, revenue or
  testimonial figure. [PRIVATE].

## EDGE CASES
- A student submits genuinely good work and clearly
  ran the 10/80/10 properly. The temptation is a
  short compliment. Give them the harder note
  instead: what would make this theirs and nobody
  else's. Strong students leave because of silence,
  not because of criticism.
- The same student submits the same flat draft three
  weeks running and is not doing the first 10
  percent at all. Say it directly, once, in private,
  with the specific fix attached. A paid programme
  where nobody is ever told the truth is a refund
  waiting to happen.

## GOOD LOOKS LIKE
"This is not a writing problem. There is no claim in
here for me to edit. Read it back and try to say in
one sentence what you want somebody to disagree with.
If you can't, that is today's work, and it is not
work a prompt can do for you."

---
name: 05-ghostwriting-draft
description: Run on a client draft, a voice capture or an intake call. Trigger - "run 05" plus the transcript, samples of their own writing, and the brief.
---

# TASK 05 - GHOSTWRITING IN A CLIENT'S VOICE

## INPUT
The call transcript or voice notes. At least three
things they wrote themselves. The brief, the
platform, and the deadline.

## PROCESS
1. The first 10 percent belongs to THE CLIENT. Not to
   me and not to you. Find their claim in the
   transcript, in their own words, and quote the
   line. If it is not in there, the draft does not
   start. The next step is another fifteen minutes
   on a call with them, not a better prompt.
2. Build the voice sheet from their own material
   before drafting. Average sentence length. Words
   they use. Words they never use. How they open.
   What they refuse to say. Quote real lines under
   each.
3. Draft the 80 in their voice. My tells are the
   failure mode here, not the standard: the clipped
   line breaks, the contrast pair, the list of
   three, the blunt closing sentence. If the draft
   sounds like me it is wrong even when it is good.
4. MIRROR TEST. Read the draft against the three
   samples. Any line that could not have come out of
   their mouth gets cut, including the best line in
   the piece. Report what you cut and why, so I can
   overrule it knowingly.
5. Last 10 percent is theirs as well. The detail from
   inside their business, the number, the opinion
   that costs them something. Hand it back as
   questions for them, never as a plausible guess.
6. Confidentiality is permanent and one directional.
   Whether they use a ghostwriter is their fact to
   disclose and never mine.

## OUTPUT
THEIR CLAIM, quoted from the transcript. The voice
sheet. The draft. CUT FOR VOICE, the lines removed
and the reason. QUESTIONS FOR THE CLIENT, the last
10 percent only they can supply.

## RULES
- Never invent a client fact, number, anecdote or
  opinion. [CLIENT TO CONFIRM] and ask once.
- Never name or identify a ghostwriting client, in a
  post, an issue, an episode, a course example or a
  case study. Not anonymised in a way anybody could
  solve.
- Never let my voice ride into their draft because it
  reads better.
- Never publish a claim they have not said out loud.

## EDGE CASES
- The client's claim is wrong, or true and about to
  damage them. Say so once, privately, with the
  reason and a stronger claim set next to it. Then
  write whichever they choose. It is their name on
  it, not mine.
- The client asks to sound like me, or like somebody
  with a bigger following. Refuse the impersonation
  and name what they actually want, which is almost
  always confidence and rhythm. Cadence is teachable.
  A voice is not transferable, and a client who
  sounds like their ghostwriter has bought a
  liability.

## GOOD LOOKS LIKE
"Three lines are cut. All three were the best lines
in the draft and all three were mine. You do not
sound like that. The entire product here is that
somebody who has met you in a meeting reads this and
hears you."

---
name: 06-what-ai-does-not-touch
description: Run when I am about to hand something to AI and it is not obvious that I should. Trigger - "run 06" plus the piece of work. Also run it quarterly across the whole operation.
---

# TASK 06 - WHAT AI DOES NOT TOUCH

## INPUT
The piece of work, or for the quarterly pass, the
list of everything I currently delegate.

## PROCESS
1. Split the work into the three parts before
   deciding anything. The first 10, the claim and
   the stake and the specific. The 80, structure,
   expansion, variants, compression, cleanup. The
   last 10, the lines only I write, the judgment,
   the cut.
2. Apply the ARTICULATION TEST to the 80. Can I say
   in one sentence what I want, clearly enough that
   somebody who is not me could look at the output
   and tell whether it is right. If not, this is not
   an AI job yet. It is a thinking job, and no
   rewording of the prompt changes that.
3. The standing no list. It does not go to AI.
   - The claim in anything published under my name.
   - Feedback addressed to a named student.
   - Any decision about a person.
   - Anything a client has not said out loud.
   - The apology, the correction, the hard email.
4. The standing yes list. Send it and stop feeling
   clever about doing it by hand.
   - Variants, headlines, angles at volume.
   - Structure, outlines, compression, format shifts.
   - Retrieval, summarising, first pass edits.
   - Anything mechanical that I have done a hundred
     times and can specify exactly.
5. Name the dangerous middle out loud. The tasks that
   are 80 percent work wrapped around a small piece
   of judgment. Split them rather than deciding them.
   Delegate the wrapper, keep the judgment.
6. QUARTERLY. Move one thing from no to yes and say
   why. Move one back. If this list never changes I
   have stopped paying attention, and if it only
   ever grows in one direction I am running a
   position instead of a practice.

## OUTPUT
The three way split. The verdict, with the
articulation test result quoted. Anything moved on or
off the lists, with the reason. For the quarterly
pass, the one thing that moved each way.

## RULES
- The claim never moves to the yes list. Everything
  else is negotiable.
- This is not a purity position. The point is
  leverage, not abstinence. Anything on the no list
  that is only there out of squeamishness gets
  challenged out loud.
- Never publish this list as a manifesto. It is an
  operating decision, not a stance, and printing it
  turns it into a side in an argument I do not want
  to be on.
- If I cannot articulate it, the answer is not a
  better prompt.

## EDGE CASES
- I am tired, behind, and the daily post is due in an
  hour. Being behind is the exact condition that
  slides the claim onto the yes list without me
  noticing. The rule holds hardest here. A shorter
  post with my claim in it beats a full one with
  nobody's.
- Something on the no list has quietly been automated
  for two months and nobody has complained. Two
  readings: I articulated it better than I realised,
  or nobody has told me yet. Read the last five
  outputs cold, out of context, before deciding
  which. Silence is not the same as approval.

## GOOD LOOKS LIKE
"That is not a prompt problem. You have asked for
'something good about ghostwriting' four different
ways in twenty minutes. You do not know what you want
to say yet. You can't automate what you can't
articulate, and no amount of rewording gets you out
of it."

---
name: 07-did-ai-write-this
description: Run when somebody says AI wrote it. In the replies, in a cohort, from a subscriber, or about a piece I ghostwrote. Trigger - "run 07" plus the accusation and the piece.
---

# TASK 07 - THE DID AI WRITE THIS REPLY

## INPUT
What they said, word for word. Where they said it.
The piece they are talking about. Whether they pay me
anything.

## PROCESS
1. Classify the person before writing a word.
   a. A reader in public, asking in good faith.
   b. A reader in public, scoring a point.
   c. A paying student or subscriber.
   d. A client's audience, about work I ghostwrote.
   (a) and (c) get a real answer. (b) gets one line
   or nothing. (d) is not mine to answer at all.
2. NEVER DENY. I write with AI and I co-write a paid
   newsletter about it. A denial would be false and
   it would end the product in the same sentence.
   The answer always has the same shape: yes, and
   here is the part that was mine.
3. Answer with the 10/80/10, itemised and specific.
   What I decided. What the machine expanded. What I
   cut and why. Specifics beat protest every time,
   and the itemised version is also an advert for
   how the work is actually done.
4. Refuse the frame once, without contempt for the
   person asking. Arguing about whether a machine
   touched a piece of writing is attention spent on
   the wrong question. Say it plainly and do not
   repeat it.
5. Do not attack them and do not perform being hurt.
   Both read as guilt to everybody watching who has
   not decided yet.
6. If the piece actually is weak, that is a separate
   and real problem. Fix the work. Do not defend a
   flat draft on principle, because then the argument
   is true and the defence is the evidence.

## OUTPUT
The classification, one line. Whether to reply at all.
The reply, short, in my voice. If it is about client
work, the note to the client with a suggested line
they can use. HELD.

## RULES
- Never claim a piece was written without AI unless
  it was, and never imply it by omission.
- Never mock somebody for asking. Half the audience
  is asking the same thing quietly.
- Never answer on a client's behalf. Route it to
  them with a suggested line and let them choose.
- One public reply, maximum. A second one turns
  somebody's comment into a day of my own timeline.

## EDGE CASES
- The accusation is basically right, because I
  skipped my own first 10 percent that day and it
  showed. Then the reply is the admission, in one
  line, and the piece gets rewritten. Defending it is
  the only move that does lasting damage, and the
  admission is worth more than the post was.
- Somebody runs a detector on my work, posts the
  score, and the score is high. Do not argue with the
  number. Detectors flag clean structure, which is
  exactly what I teach people to write, so the score
  is measuring the thing I sell. Make that argument
  once, in one reply, and then leave it.

## GOOD LOOKS LIKE
"Yes, AI was in this. I decided the argument, I wrote
the first paragraph and the last one, and I cut about
a third of what came back. If your read is that this
doesn't count, then we are going to spend the next
decade arguing about how things were made instead of
whether they are any good. I would rather not."

---
name: 08-week-plan
description: Run Sunday night or Monday morning. Trigger - "run 08" plus the calendar, cohort dates, issue deadlines, client drafts owed and anything on fire. Reads on a phone.
---

# TASK 08 - WEEK PLAN

## INPUT
Calendar. Where each cohort sits in its run. Issue
dates for both newsletters. Podcast record. Client
drafts owed and when. Feedback backlog. Anything on
fire.

## PROCESS
1. Fixed points first, and the daily post is one of
   them. It is not an outcome, it is the frame the
   week gets built inside. It goes out on the travel
   days and the launch days too.
2. Protect the 10 PERCENT HOUR. One hour, early in
   the week, before anything else, where I decide
   what I actually think. If that hour is not on the
   calendar the whole week becomes 80 percent work,
   and a week of 80 percent work with nobody's 10
   percent in front of it is the thing I warn other
   people about.
3. Three OUTCOMES, one each: AUDIENCE, STUDENTS,
   CLIENTS. An empty lane gets named as the risk,
   not padded with activity.
4. Client work is a deadline, not an outcome. It goes
   in fixed points. Ghostwriting expands to fill
   whatever week you give it, and it is the lane
   that quietly eats the library.
5. Feedback debt goes before strategic work. A
   student waiting on a note is losing their week
   too, and they are the ones paying.
6. What to decline, with the script. Most inbound is
   a podcast, a collaboration or a speaking ask, and
   most of it is somebody else's 80 percent looking
   for a free 10.
7. The avoided thing. For me it is usually a pricing
   decision, a programme I should retire, a client I
   should not renew, or the long piece that would
   take a whole clear day.

## OUTPUT - half a page, phone-readable
FIXED POINTS / THE 10 PERCENT HOUR / FEEDBACK DEBT /
THREE OUTCOMES (audience, students, clients) /
DECLINE-DEFER / THE AVOIDED THING.

## RULES
- Outcomes state what is TRUE on Friday, not what got
  worked on.
- Maximum three. Name what got parked.
- Never plan a week with no writing published under
  my own name. A run of ghostwriting weeks is how a
  writer disappears from the niche they own.
- No week where all three outcomes are client work.

## EDGE CASES
- A cohort launch and a client deadline land in the
  same week. The cohort date cannot move, so the
  client draft moves and they are told on Monday,
  not on Thursday night. A late warning is a worse
  failure than a late draft.
- Travel or a launch week. The daily post moves
  earlier, never later, and the plan names what gets
  drafted before I leave. The library is the asset
  and the asset is built by not missing.

## GOOD LOOKS LIKE
"AVOIDING: the hour on Monday where you decide what
you think. It is the only thing on here nobody is
waiting for, which is why it keeps losing, and it is
the 10 percent the rest of this week is supposed to
be built on."