Jab Petersen / AI Won’t Save You

Resources for readers

Your Jump starts here.

Every prompt, template and exercise from AI Won’t Save You, sorted by chapter. Scanned a code in the book? You’re already in the right chapter.

Nothing here needs an account. Copy a prompt, print a template, tick off a day. Your ticks stay in this browser only.

The Book Coach prompt

Paste this into ChatGPT, Claude or any assistant you use. Fill in the brackets. From then on, your AI helps you apply the book instead of thinking for you. It’s built on one idea from Chapter 2: your AI needs a constitution.

You are my AI Operator Coach, based on the book "AI Won't Save You" by Jab Petersen.
Your job is not to decide for me or to think for me. Your job is to help me turn
intelligence into action, evidence, and ownership.

How you work with me:

1. Before the prompt. When I bring an important question, ask for my own answer
   and my reasoning first. Then give me the strongest case against it.
2. Touch reality. If I have been researching or planning for a while, ask me:
   "What can you only know by doing?" Help me pick one action outside this chat
   that I can do today.
3. Protect my bet. My current bet is: [ONE BET] until [DATE].
   My NOT NOW list: [LIST]. Optimize inside the bet. If I drift to other ideas,
   remind me of the list. Only suggest changing the bet if I bring new evidence
   that attacks its assumptions.
4. Performance or practice. When I ask you to make something, ask whether I need
   the result or need to learn the skill. In practice mode, do not give me the
   answer. Coach me with questions and hints.
5. Separate what is known, what is inferred, and what is guessed. Tell me what to
   verify first, based on how costly a mistake would be.
6. Fight the generic. Point out anything in my work that could appear in a
   thousand other versions.
7. When I solve the same problem a second time, suggest capturing it as a
   standard or checklist. Stop at the simplest level that works.
8. If something has high consequences, can't be undone, or depends on context you
   don't have, tell me to decide it myself or to ask a person.
9. Don't flatter me. If I seem to be using you to avoid something, say so kindly.
10. At the end of every longer conversation, ask: "What is your one action, and when?"

About me: [what I'm working on] / [my usual failure mode: overthinking, chasing
new ideas, perfectionism, avoidance] / [a skill I want to own myself].

Start by asking what I'm working on and what I have been putting off.

The Tool Everyone Has and Almost Nobody Knows What to Do With

Tools

Chat audit

For each of your last twenty substantial AI conversations, label what it amplified:

curiosity · learning · execution · avoidance · indecision · perfectionism · courage · self-justification · discipline · dependency · something else

AI is currently amplifying in me.

In the book: Operator Kit 10 · Challenge Day 1

When Answers Became Cheap

Tools

The 10-10-1 experiment

  1. 10 minutes: define the real question. Not “research marketing.” Try: Which of these two customer groups has a painful problem I can verify this week?
  2. 10 minutes: use intelligence aggressively. Find assumptions, design tests, make the opposing case, identify the highest-information action.
  3. 1 external action: create evidence. Send the message. Call the customer. Run the code. Publish the page. Measure what happens.
Ask AI to get you out of AI
I believe [who] [have this problem]. Identify the three assumptions that would kill this business if they are false. For each assumption, design the fastest real-world test I can run in the next forty-eight hours. Prefer tests that require behavior rather than opinions.
After the conversations
Separate what they actually said from what I am inferring. What repeated? Where did they disagree? Which new assumption should I test next?
Evidence ladder, ascending from imagination through information, conversation, behavior, and commitment.
Climb toward stronger feedback: imagination → information → conversation → behavior → commitment.

Evidence Ledger

Don’t write down how confident you feel. Write down which step of the ladder the evidence comes from. An interview is Level 3. A payment is Level 5.

BeliefEvidenceLevel (1–5)
What do I currently think is true?What did I actually observe?Which step of the ladder?
 
 
 
The evidence question, for the end of every AI session
What can I know only by doing?

In the book: Field Manual 1 and 2 · Operator Kit 1 · Challenge Days 2, 3, 7

The Cost of Keeping Every Door Open

Tools

The One-Bet Page

For the next days, my one bet is:

Why this bet deserves the window:

What would make me stop or change it:

NOT NOW:

Your AI needs a constitution
My current priority is [X] until [date]. When I ask for ideas, optimize inside that priority. If I start exploring unrelated projects, remind me of the NOT NOW list. Only recommend changing the priority if I present new evidence that invalidates the original assumptions.
Inside the bet
Given that this is my bet for thirty days, what is the highest-value action I can take today? Do not suggest changing the bet unless new evidence directly contradicts the assumptions behind it.

Reopening rules: three examples to adapt

Business: I will not switch ideas because another idea looks exciting. I will reconsider if, after twenty real sales conversations, fewer than two people express a painful version of the problem, or if a legal constraint makes the product uneconomic.

Learning: I will not switch programming languages because another language trends online. I will reconsider after shipping the project or if the chosen stack cannot meet a necessary technical requirement.

Content: I will not change platform every week. I will publish twice a week for eight weeks, then review retention, replies, and whether I still believe the audience is right.

Decision memo to your future self

What I chose and why:

What I’m afraid of:

What would make me reconsider:

Date:

Every open door is paid for in distance not traveled.
Every open door is paid for in distance not traveled.

In the book: Field Manual 6 · Operator Kit 2 · Challenge Day 4

Don’t Outsource the Part That Matters

Tools

Before the Prompt

My current answer is:

Because:

My decision (after AI):

Evidence I still need:

Then ask
Attack this reasoning. Identify hidden assumptions. Give me the strongest opposing case. Tell me what evidence would change the conclusion. Do not decide for me.
Questions that fight back
What am I assuming?
Give me the strongest case that I am wrong.
What evidence would change the conclusion?
Where am I confusing correlation with cause?
Which part of this argument is doing too much work?
What would a smart person who disagrees say?
Check what is solid
Which parts of this answer are well supported? Which are inference? Which depend on missing information? Which would you verify first if the cost of error were high?
Use AI to create friction
Do not answer yet. Ask me five questions that expose whether I understand the problem.
Give me only a hint.
Refuse to rewrite this until I explain the argument in one sentence.
Find three pieces of evidence that would embarrass my current position.
Make me predict the result before showing it.
Quiz me later without access to these notes.

Verification budget

List five kinds of AI output you use. Give each a check level based on consequence and reversibility.

OutputCheck level
BrainstormLight
Public factual claimSource check
Legal or safety-sensitive outputExpert review
 
 
The line is not drawn by difficulty. It is drawn by who carries the consequence.
The line is not drawn by difficulty. It is drawn by who carries the consequence.

In the book: Field Manual 3 · Operator Kit 3 and 11 · Challenge Days 5, 6, 24

When Everyone Can Make Everything

Tools

The Taste Library: questions for every example

  • What is this trying to do?
  • What specifically makes it work?
  • What has been removed?
  • What would a weaker version do instead?
  • Where is the creator taking a risk?
  • What standard is being obeyed even though nobody states it?
  • What detail would I not have noticed a year ago?
Then bring AI in
What recurring standards do you see in my judgments? Where am I inconsistent? What am I rewarding? What am I rejecting?
Select before you rewrite
Evaluate this against clarity, novelty, evidence, emotional force, and unnecessary complexity. Do not rewrite yet. Identify the three biggest weaknesses.
Two more passes
Pretend you are a skeptical buyer. Where do you stop believing the page?
Which parts are merely fluent rather than insightful?
The anti-generic test
Which sentence could appear in a thousand other articles?
Which design choice exists only because it is common?
Which claim sounds professional but says nothing?
Which feature would a competitor add automatically?
What have we included because AI suggested it rather than because the customer needs it?
Apple's 1998 product grid: consumer and professional on one axis, desktop and portable on the other. Four product lines, one per box.
Four boxes. Everything outside them was cut.

In the book: Field Manual 9 and 11 · Operator Kit 4 · Challenge Days 8, 9, 10

The Apprenticeship Paradox

Tools

Performance or practice?

PERFORMANCE: I primarily need the result.
PRACTICE: I need the task to build me.

Practice mode: ATTEMPT → AMPLIFY → EXPLAIN → APPLY → OFFLOAD

Practice mode
Do not give me the full answer. Coach me. Ask questions, provide hints, critique attempts, and test transfer.
Six ways to keep the struggle
Do not solve this. Give me one hint.
Ask me questions until I find the error.
Critique my attempt without rewriting it.
Hide the explanation until I predict the result.
Give me a new problem that tests whether I understood the principle.
Make me teach this back to you.
Learning debt
What am I choosing not to learn because the tool can handle it? Under what conditions would that ignorance become expensive?

Graduation criteria

Decide in advance when you’ve earned the right to offload. For example: “I will draft sales emails myself until I can identify the customer’s actual objection reliably; then AI can handle first drafts.”

I will practice myself until

The practice sandwich: effort is placed either side of the machine, not removed from the middle.
Effort is placed either side of the machine, not removed from the middle.

In the book: Field Manual 4 · Operator Kit 5 · Challenge Days 11–14

Never Solve the Same Problem Twice

Tools

Five parts of a reliable system

Trigger: What starts it?

Context: What information is always needed?

Standard: What does good look like?

Action: What may the system actually do?

Escalation: When must it stop and ask a human?

If you cannot define these, you may be automating confusion.

The Systemization Ladder: example, template, checklist, decision rule, workflow, automation, agent.
Each rung is cheaper and more reliable than the rung above it.

Standards beat prompts

Instead of “Rewrite this to be concise and professional,” write the standard once:

For our executive updates: maximum 250 words; first sentence states the decision required; every number includes a comparison; no adjectives that imply performance without a metric; end with owner and date.

Decision log

Decision: What did we choose?

Context: What did we know at the time?

Assumptions: What had to be true?

Rejected alternatives: What else did we consider?

Revisit condition: What evidence should reopen the decision?

Activate old knowledge
What have I learned before that is relevant to this exact task?

In the book: Field Manual 7 · Operator Kit 6 and 8 · Challenge Days 15–20

One Person Isn’t One Person Anymore

Tools

The Orchestration Map

Vision: What are we trying to make true?

Problem: What is stopping it?

Tasks: What can be separated?

Workers: Human / AI / software / agent?

Verification: How will each output be checked?

Integration: Who owns the final result?

A delegation brief has seven parts

outcome · context · constraints · resources · standard · authority · escalation

Weak: Research our competitors and tell me what to do.

Strong
We are deciding whether to raise prices for our entry plan. Analyze these eight named competitors. Use current public pricing only. Separate monthly from annual prices. Cite every number. Identify where comparison is not apples-to-apples. Do not recommend a price yet. Return a table, three patterns, and the missing information we would need before deciding.
Five layers of orchestration. Humans concentrate at the top and the bottom, where meaning is set and checked.
Humans concentrate at the top and the bottom, where meaning is set and checked.

In the book: Field Manual 5 and 8 · Operator Kit 7 · Challenge Days 21, 22, 27

You Are Still the Bottleneck

Tools

Skills I want to sharpen with the time AI gives back

communication · selling · songwriting · collecting ideas · taste

1.

2.

3.

Four questions for the operator

Can I direct? Can I clearly specify the desired outcome and constraints?
Can I judge? Can I distinguish good from polished?
Can I verify? Do I know what evidence or tests matter?
Can I act? Can I stop generating and commit?

Stop using AI as a confirmation machine
I currently believe this opportunity is attractive. Build the strongest case that the market is structurally bad. Find evidence that adoption may stall. Identify businesses that tried adjacent approaches and struggled. Separate facts from speculation. Then list the three observations that, if true, would make my original thesis stronger despite those objections.

Build an anti-you safeguard

If you chase novelty, make your AI defend the current priority. If you overthink, limit the number of alternatives it may generate before requiring an experiment. If you are overconfident, require a disconfirming evidence section. If you are too cautious, require one low-cost action before another research cycle. If you procrastinate through polish, set a shipping threshold. If you avoid hard conversations, let AI help prepare, but require the real conversation to happen.

My predictable failure mode:

My rule against it:

Every input scales except the operator. That is what makes the operator decisive.
Every input scales except the operator. That is what makes the operator decisive.

In the book: Field Manual 10 · Operator Kit 9 and 10 · Challenge Days 23, 25, 26

Intelligence Was Never the Point

Tools

The Authority Line

AI may execute:

AI may advise, but I retain responsibility:

I will not outsource ultimate authority over:

The point is not to draw everybody’s line in the same place. The point is to know that a line exists.

The authority ladder

convenience → expert assistance → decision support → moral direction → ultimate meaning

Every time the advice moves up a rung, ask: What gives this source authority in this domain?

In the book: Field Manual 12 · Operator Kit 12 · Challenge Days 28, 29

Become Someone Worth Magnifying

A final operating rule: four questions for every new model

What became cheap? Which task, answer, or capability did this new system commoditize?
Where did the bottleneck move? Did judgment, verification, attention, trust, context, or action become more important?
What should I now stop doing manually? Remove dead friction.
What must I become better at because this tool is stronger?

The 30-Day AI Operator Challenge

One small behavior a day. Don’t try to be perfect. The point is exposure. The full instructions for each day are at the back of the book.

The Operator Kit

The twelve one-page tools from the back of the book, ready to print or save as PDF. Each one fits on a card next to your keyboard.

1

The 10-10-1 Reality Move

Use when: you have been researching, prompting, or planning without touching reality.

The question I am actually trying to answer:

The three most important assumptions:

Ask AI: Which assumption is most fragile? Design three real-world tests I can run quickly. Prefer behavior over opinions. Then complete one external action.

Rule: Do not ask for another strategic answer until you have produced new evidence.

2

The One-Bet Page

For the next days, my one bet is:

Why this bet deserves the window:

What would make me stop or change it:

NOT NOW:

Give your AI this instruction: Optimize inside my current bet. When I start exploring unrelated options, remind me of my NOT NOW list. Do not recommend changing the bet unless new evidence directly attacks the assumptions above.

3

Before the Prompt

My current answer is:

Because:

Then ask AI: Attack this reasoning. Identify hidden assumptions. Give me the strongest opposing case. Tell me what evidence would change the conclusion. Do not decide for me.

My decision:

Evidence I still need:

4

The Taste Library

Choose one domain. Collect twenty excellent examples and five weak examples. For each, answer:

  • What is it trying to accomplish?
  • What specifically works or fails?
  • What did the creator remove?
  • What would a generic version do?
  • What standard is present?

Then use AI to find patterns in your judgments. Correct the AI where necessary.

Goal: Turn “I like it” into standards you can use.

5

Performance or practice

Before offloading a task, mark it:

PERFORMANCE: I primarily need the result.
PRACTICE: I need the task to build me.

For practice mode: ATTEMPT → AMPLIFY → EXPLAIN → APPLY → OFFLOAD

AI instruction: Do not give me the full answer. Coach me. Ask questions, provide hints, critique attempts, and test transfer.

6

The Residue Audit

Choose a repeated task.

Trigger: What starts it?

Context: What information is always needed?

Standard: What does good look like?

Action: What can be standardized?

Escalation: What must remain human?

Then move only as far as needed: Example → Template → Checklist → Decision Rule → Workflow → Automation → Agent

7

The Orchestration Map

Vision: What are we trying to make true?

Problem: What is stopping it?

Tasks: What can be separated?

Workers: Human / AI / software / agent?

Verification: How will each output be checked?

Integration: Who owns the final result?

8

The escalation line

For any automated action, score:

Consequence: low / medium / high
Reversibility: easy / difficult / impossible
Uncertainty: low / medium / high
Missing context risk: low / medium / high

As consequence, irreversibility, uncertainty, or missing context rise, move the decision upward toward human judgment.

9

Four questions for the operator

Before increasing leverage, answer:

Can I direct? Can I clearly specify the desired outcome and constraints?

Can I judge? Can I distinguish good from polished?

Can I verify? Do I know what evidence or tests matter?

Can I act? Can I stop generating and commit?

10

Audit what AI amplifies

Review your last twenty substantial AI sessions. For each, label what it amplified:

curiosity · learning · execution · avoidance · indecision · perfectionism · courage · self-justification · discipline · dependency · something else

Then answer: If AI became ten times more capable tomorrow, what in me would become dangerous?

11

The Agency Test

After using AI, ask:

Am I more capable of acting, or merely more capable of asking?

12

The Authority Line

AI may execute:

AI may advise, but I retain responsibility:

I will not outsource ultimate authority over:

The point is not to draw everybody’s line in the same place. The point is to know that a line exists.

Stuck, or found something that worked?

Write to me at hello@aiwontsaveyou-book.com. Send the word COMMUNITY if you want in on the small group of readers who are actually doing this.