• Gall’s Law is the pattern that complex things that work usually started as simple things that worked.

• Most personal systems fail because they depend on too many things going right, every day.

• The practical loop is Minimum → Run → Learn → Upgrade (or Quit).

• AI helps by shrinking your plan, spotting fragile dependencies, and proposing one safe upgrade at a time.

• After reading, you will have three copy paste prompts to turn any messy goal into a small working routine.

In five minutes, you can turn “I need a full system” into a loop you will actually keep.

1) Why big plans feel good (and why they die)

You decide you will finally fix it.

Health. Focus. Home. Relationships. Pick your arena.

You do what smart people do. You think.

You research, compare options, and design a complete plan. You add rules so you will not fail. You collect tools so it feels serious.

For a moment, it feels great. Your brain relaxes. You have a map.

Then real life shows up.

A bad night of sleep. A busy day. One missed day.

And the plan needs momentum.

So one missed day becomes two. The plan starts to feel heavy. You stop using it.

Then the story becomes, “I just lack discipline.”

Most of the time, the system was simply too demanding. It needed your best self too often.

This is a way to change that does not depend on motivation. It depends on design.

I know this trap well. I tend to overthink. I used to believe I could think a problem to completion and then execute cleanly. What changed things was a bit of indifference. I started with a small imperfect version and let reality teach me what to improve next.

This week’s move is simple:

Build the smallest loop that survives your worst day.

How AI helps

AI gives you speed and clarity:

• It shrinks your plan into a minimum you can keep.

• It shows what your plan depends on.

• It helps you upgrade without restarting.

Key takeaway: A plan that requires perfect days will not survive normal life.

2) Gall’s Law in plain language (and why it works)

Gall’s Law is a simple pattern.

If you want something complex that works in real life, it usually started as something simple that worked.

This is about building something that survives contact with normal days.

The hidden problem

Most full systems fail because they depend on too many things going right at the same time.

Your plan quietly depends on sleep, mood, time, prep, motivation, and a clean week.

When any one of those breaks, the whole thing breaks.

A simple start cuts the dependency chain.

It gives you one small loop you can keep running long enough to learn.

The fitness example

People chase the perfect plan, the perfect program, the perfect gear.

A better start is one exercise.

Day 1: 1 rep.
Days 2 to 3: 2 reps.
Days 4 to 6: 3 reps.

You are keeping the loop alive.
Technique improves because you repeat it.
Strength improves because the load grows slowly.
Habit forms because the entry cost stays tiny.

The practical move

Gall’s Law gives you three options when you want change:

• Start with a minimum loop

• Run it until reality shows friction

• Upgrade one thing at a time

If the loop never produces a signal, stop and choose a different loop.

If you liked our Occam’s Razor and Via Negativa articles, this is the same instinct in action: simplify, remove friction, then grow what earns its place.

Quick source note: this principle is commonly attributed to John Gall’s Systemantics (1970s).

Key takeaway: Start small. Keep it alive. Let reality choose the complexity.

Now we turn the model into a five minute protocol you can run on demand.

3) Prompts: use AI to run Minimum → Run → Learn → Upgrade

Gall’s Law is easy to agree with and easy to ignore.

So the goal is speed. You want a small loop you can run in minutes, right when you feel the urge to design a full system.

Pick the prompt that matches your stage. Do not run all of them.

If you only do one, do 3.1.

3.1 Shrink prompt: define the Minimum Working Version (Minimum)

Use this when your goal feels big and your brain wants a complete plan.
You will know it worked if you can do the minimum on a bad day and you still did it.

Help me apply Gall’s Law.

Return the answer in three blocks:
A) Minimum Working Version (under 2 minutes)
B) Success signal (one simple signal the loop is alive)
C) Anti creep (top 3 things to NOT add yet)

Then:
1) Make the minimum doable on my worst day.
2) Write the one sentence rule I will follow for 7 days.

INPUT:
Goal + constraints (time, energy, schedule, injuries, budget, preferences):
[PASTE HERE]

3.2 Diagnostic prompt: find the fragile dependencies (Learn)

Use this when you already have a plan, but it keeps collapsing for reasons you cannot name.
You will know it worked if you can name the top 3 breakpoints and one subtraction you will try today.

Audit my routine for fragility.

Return the answer as a short checklist with four sections:
1) Top 3 must go right dependencies
2) One way to weaken each dependency
3) The smallest change I can apply today
4) A fallback minimum mode for bad days

INPUT:
Routine + where it breaks (what happens, what I skip, what day it tends to fail):
[PASTE HERE]

3.3 Decision prompt: upgrade without restarting (Upgrade or Quit)

Use this when the loop is alive and you want better results without breaking consistency.
You will know it worked if you have one upgrade, one stop rule, and minimum mode stays alive.

Help me evolve a working loop without restarting it.

Return answers in five parts:
1) What reality taught me (constraints, friction, failure points)
2) ONE reversible upgrade for the next 14 days
3) One metric (what "better" means)
4) One stop rule and one fallback (minimum mode stays alive)
5) Decision: keep, upgrade, downgrade, or quit
Then write the next 14 days rule in one sentence.

INPUT:
Current loop + last 7–14 days (what I did, what I missed, what felt hard, any results):
[PASTE HERE]

Prompts are the tactic. The next section is the operating system.

4) Principles and traps

This is not a motivation problem. It is a design problem.

Principles

• Keep the loop alive.
  Your minimum is sacred.

• Fewer dependencies win.
  Build something that does not require perfect days.

• One change at a time.
  Change one thing, learn one thing.

• Prefer reversible upgrades.
  Improve without breaking what already works.

Notes that help:

• Let reality choose the complexity.

• Use AI to subtract steps and decisions.

Traps

• High stakes blindness.
  Iterating casually when the downside is meaningful or irreversible.

• System cosplay.
  Tools, templates, gear, research. It feels like progress. Nothing runs.

• Restart addiction.
  Missing once turns into “start over Monday.”

• Complexity creep.
  Every improvement adds a step until the whole thing collapses.

• Premature optimization.
  Solving problems you do not have yet.

• AI theater.
  Beautiful plans that you do not execute.

5) Closing: the practical payoff

You do not need a perfect system.

You need a loop that survives real life.

The next time you feel the urge to design the complete plan, run the loop:

Minimum → Run → Learn → Upgrade (or Quit)

Minimum: What is the smallest version I will actually do.
Run: Can I keep it alive for seven days.
Learn: What broke, and what does that reveal.
Upgrade: What one reversible change improves it.
Quit: If there is no signal, stop and pick a different loop.

If you want more pieces like this, subscribe on Substack. One mental model per week, with prompts you can use immediately.

If you want, paste the goal you are overbuilding right now in the comments and I will help you find the Minimum Working Version.

• Survivorship bias is when you learn from the visible winners while the failures quietly disappear.

• It makes success look more common and more repeatable than it really is.

• The practical loop is Filter → Graveyard → Recompute → Low-regret action.

• AI helps by surfacing hidden filters and turning winner advice into conditional truth.

• After reading, you will have three copy-paste prompts to turn any success story into a reversible test.

In five minutes, you can turn “this worked for me” into a small, safe experiment with a stop rule.

1) Why winner advice hooks you (and why it burns at times)

You see a transformation post.

“I lost 12kg in 8 weeks. Here’s the diet. It’s simple.”

Then you try it.

Week three: you are hungry and irritable. Social life gets awkward. You slip once, then twice, and the story in your head becomes, “I just lack discipline.”

That is the trap. The post did not show you the full distribution of outcomes.

Winners stay visible. Everyone else quietly disappears.

So you start treating “worked for them” as “works.”

This week’s move is simple:

Before you copy the advice, restore the denominator.

How AI helps

AI cannot reveal the true denominator.
But it can force the right questions before you commit:

• What had to be true for this story to be visible to me?

• Who disappears when it fails?

• What would I do differently if success is much rarer than it looks?

Key takeaway: Winner stories are not lies. They are incomplete samples.

2) Survivorship Bias in plain language

Survivorship bias is when you draw conclusions from what survived a filter, while what failed is missing.

Here is the main mechanism: when visibility depends on success, your picture of reality becomes systematically too optimistic.

The Diet Trap

Diet success stories are useful, and they are distorted.

Useful, because they show what a result can look like.
Distorted, because they hide how many people quit, did not respond, or dropped out for reasons that never make it into the story.

So the conclusion you want is not “this works.”

It is:

For people who could follow it and tolerate it, it produced results.

That rewrite removes the magic.
It turns the question from “Does it work?” into “What does it require, and can I meet that consistently?”

Not the same as…

Selection bias is a broad family of problems where what you observe is not representative.
Reporting bias is when wins get shared more than losses.
Survivorship bias is the specific case where the losers are missing because they did not make it through the filter, or they stopped being measured.

Define your question first

If your question is “What do successful people tend to do?” winners can help.
If your question is “How likely is this to work for me?” you need the denominator.

Key takeaway: Convert winner advice into a conditional statement before you act.

Now that you have the mechanism, here is the three-prompt loop to run it under time pressure.

3) Prompts: use AI to run Filter → Graveyard → Recompute → Decide

Survivorship bias is easy to understand and easy to ignore.
So the goal is speed. You want a small loop you can run in minutes, right when you feel the urge to copy a winner.

Pick the prompt that matches your stage.
Do not run all of them.

If you have medical conditions or an eating disorder history, do not experiment with diets without professional guidance.

3.1 Diagnostic prompt: identify the filter

Use this when you see a success story and feel pulled to imitate it.

Tip on what to paste: paste the post plus your constraints (schedule, preferences, history, injuries, budget).

I want to check for survivorship bias in this advice.

Return answers in three buckets:
A) Known (explicitly supported by what I pasted)
B) Inferred (reasonable, but not stated)
C) Guess (speculation, label it)

1) Summarize the claim in one sentence.
2) What had to be true for this story to be visible to me?
3) Who is systematically missing from what I am seeing?
4) What would be the most common reasons those missing cases disappear?
5) Rewrite the claim as a conditional statement (not universal).
6) What one question would most reduce uncertainty?

STORY / ADVICE:
[PASTE HERE]

3.2 Reconstruction prompt: list the invisible graveyard

Use this when you want to act, but you only have winner evidence.

This is not about guessing numbers.
It is about naming the missing buckets so you do not hallucinate certainty.

Tip on what to paste: paste the method plus your real-life constraints.

Help me reconstruct the missing denominator behind this success story.

1) List the main categories of people who tried this but did not get the outcome.
2) For each category, give the most likely reason they disappear from view.
3) For each category, propose one simple way I could look for evidence
   (reviews, dropout discussions, long-term follow-ups, base rates, etc).
4) What would be a conservative assumption about success rate
   if I include these missing cases?
5) What are the biggest unknowns that could flip the conclusion?

STORY / METHOD / MY CONTEXT (constraints, lifestyle, health, time, budget)::
[PASTE HERE]

3.3 Decision prompt: choose a low-regret next step

Use this when you want to try something without overcommitting.

Tip on what to paste: paste the advice plus what “success” means for you.

Turn this winner advice into a low-regret plan.

1) Rewrite the claim as conditional truth.
2) List what it requires to work (time, adherence, tolerance, tradeoffs).
3) Design a 14-day test that is reversible:
   - the smallest version I can try
   - one success metric
   - one stop rule
4) Give me a safer fallback option if adherence fails.
5) Tell me what would count as "evidence it is not for me."

ADVICE / CONTEXT:
[PASTE HERE]

Prompts are the tactic. The next section is the operating system.

4) Principles and traps

Survivorship bias is not a math problem.
It is a discipline problem.

You will keep forgetting the denominator unless you install a few rules of thumb.

Principles

• Winner stories are conditional evidence.
  They can teach you what worked for someone.
  They cannot tell you how often it works, or for whom.

• Separate three questions.

  1. Does it work in principle?

  2. Can people stick to it?

  3. Does it fit my life?
     Most advice collapses all three into one confident sentence.

• Think in ranges, not certainty.
  You rarely need the true denominator.
  You just need to avoid the fantasy that success is common.

• Match commitment to downside.
  If the downside is small, test quickly.
  If the downside is meaningful, demand stronger evidence and tighter stop rules.

• Prefer reversible tests.
  A good trial is one you can stop without damage.
  That is how you learn without betting your identity on a story.

Traps

• Hero worship.
  Treating one survivor as proof of a universal law.

• Cynicism.
  Going from “winner stories mislead” to “nothing matters.”
  The goal is better decisions, not despair.

• Denominator theatre.
  Pretending you can fully measure the graveyard. You usually cannot.
  Name the missing buckets and act conservatively.

• Outcome copying.
  Copying the visible result instead of the hidden requirements.

• AI overreach.
  Letting AI generate confident narratives about unseen failures.

Use it to produce hypotheses and questions, not certainty.

5) Closing: the practical payoff

Better sampling turns inspiration into good decisions.

The next time you see a story that feels like a shortcut, run the loop:

Filter → Graveyard → Recompute → Low-regret action

• Filter: What had to be true for this to be visible?

• Graveyard: Who is missing from what I am seeing?

• Recompute: What is true among survivors, and what is true in general?

• Low-regret action: What is the smallest reversible test that still makes sense if success is rare?

Your feed is not a dataset. Your life is.

Before you commit to the next diet, routine, investment, or life rule:

Respect the denominator.

If you want, paste a winner story you are tempted by in the comments and I will help reconstruct the denominator.

• Chesterton’s Fence is a simple rule for life changes: understand why a rule exists before removing it.

• It protects you from a common loop: remove “annoying” structure, feel better for a moment, then the old problem returns.

• It does not mean “old rules are good.” It means “don’t change what you don’t understand.”

• The practical loop is Recover → Upgrade → Confirm.

• AI helps you do this fast by recovering the missing “why,” stress-testing consequences, and suggesting better replacements.

After reading, you will have three copy-paste prompts to decide: keep, upgrade, or remove a rule without regret.

1) Smarter simplification in real life with AI

You remove a rule because it feels annoying.

• “No screens before bed.”

• “Always keep an emergency buffer.”

• “No heavy talks when we’re tired.”

I have broken enough “annoying” rules to learn that some of them were guarding something real.

Sleep gets worse.
Money leaks.
Small conflicts turn into big ones.

That is the trap: you did not remove a rule. You removed a protection.

So the real question is:

What problem was this rule solving, and do I still need that protection today?

That question is the heart of this week’s mental model: Chesterton’s Fence.

How AI helps

AI helps you run a simple loop in minutes:

• Recover the missing “why” behind the rule.

• Upgrade it so you keep the protection with less friction.

• Confirm with a small reversible test instead of guessing.

That is the goal here: simplify rules without accidentally repeating old mistakes.

2) Chesterton’s Fence in plain language and why it works

Chesterton’s image is simple.

You see a fence across a road.

If you do not know why it is there, you are not ready to remove it yet.

Not because fences are sacred.

Because the fence might be preventing something you cannot see from where you stand.

A fence is any default that creates friction now to prevent pain later.

In everyday life, a “fence” can be:

• a habit (“no screens before bed”)

• a money rule (“keep a buffer”)

• a relationship rule (“pause arguments when tired”)

• a family rule (“no phones at dinner”)

The common mistake is to treat friction as proof the rule is dumb.

Friction is often the receipt for a problem you forgot you solved.

The hidden pit

Most fences exist because there is a pit nearby.

You may not remember the pit.
You may not even know it happened.
But the fence stayed.

Remove the fence without finding the pit, and you learn the reason the hard way.

So the better question is not “Do I like this rule?”

It is:

What bad outcome does this prevent?

Two equal and opposite mistakes

This model protects you from two errors:

1. Blind removal
   You remove habits because they feel restrictive, then you reintroduce an old problem.

2. Blind respect
   You keep a rule forever because it is old, even when it no longer helps.

So the goal is not “keep the fence.”

The goal is:

Understand the purpose, then decide.

The practical move

Once you know the purpose, you have three options:

• Keep it

• Upgrade it

• Remove it

Simple, but powerful.

3) Prompts: use AI to Recover → Upgrade → Confirm

Chesterton’s Fence becomes useful when you can apply it quickly.

Pick the stage you are in:

If you feel friction but do not know why → 3.1.
If you know why but hate the form → 3.2.
If you tested a change and have notes → 3.3.

Do not run all prompts at once.

3.1 Diagnostic prompt: recover the hidden purpose (Recover)

Use this when you want to remove a rule but you are not sure what it was protecting you from.

Apply Chesterton’s Fence to this rule.

Return answers in three buckets:
A) Known (explicitly supported by what I pasted)
B) Inferred (reasonable, but not stated)
C) Guess (speculation, label it)

1) State the rule in one sentence.
2) What problem was it likely designed to solve?
3) What “hidden pit” does it prevent (the failure mode)?
4) What signs suggest the rule still matters today?
5) What signs suggest it is outdated or harmful?
6) Who benefits from keeping it, who pays the cost?
7) Recommendation: keep / upgrade / remove, and why.

CONTEXT (what the rule is, where it shows up, what happened before):
[PASTE HERE]

3.2 Upgrade prompt: keep the protection, reduce the friction (Upgrade)

Use this when the rule has a real purpose, but the current form is too rigid.

Help me upgrade this “fence” without losing what it protects.

1) Restate the protection in one sentence (the purpose).
2) List the top 3 costs of the current rule (stress, time, freedom, conflict).
3) Propose 3 upgraded versions:
   - A) strict but simple
   - B) flexible default with clear exceptions
   - C) different safeguard (replace the rule entirely)
4) For each version, name the failure mode it still prevents.
5) Pick the best version for my context and explain why.
6) Give examples of exceptions that would actually occur in my life.
7) Give me one friction fix to make it easier to follow.

MY CONTEXT:
[PASTE HERE]

3.3 Follow-up prompt: test and decide (Confirm)

Use this after you tried an upgraded rule for 7–14 days.

Help me decide if this upgraded fence should become a durable default.

1) Restate:
   - the original protection
   - the upgraded rule I tried
2) What would I expect to see if it’s working?
3) Summarize the evidence I observed (not vibes).
4) Decide: keep / iterate / remove, and why.
5) If keep: give me a long-term low-effort version.
6) If iterate: change ONE variable and give me the next 7-day test.
7) If remove: propose the minimum safeguard I should keep anyway.

RESULTS / NOTES & WHAT I TRIED:
[PASTE HERE]

4) Principles and traps

Chesterton’s Fence works best as a quick check, then a small experiment.

Principles

• Preserve function, not form.
  Keep the protection. Change the mechanism.

• Find the hidden pit.
  Name the failure mode the rule prevents.

• Match change size to risk.
  The higher the downside, the slower you change.

• Prefer reversible tests.
  Try a version you can undo, then decide based on evidence.

• Ask “who benefits?”
  Some fences protect people. Some protect power.

• Teach the why.
  If you want boundaries that work, the reason matters as much as the rule.

Traps

• Convenience amnesia.
  Confusing “annoying” with “useless.”

• Status quo bias.
  Using the fence as a veto on all change.

• Perfectionism.
  Waiting for certainty, then never improving anything.

• Rule-counting.
  Removing rules for the feeling of simplicity while losing safety.

• AI overreach.
  Treating AI as the authority instead of a thinking tool.

5) Closing: the practical payoff

You do not need more rules.

You need better reasons.

The next time you want to change routines, run the loop:

• Recover what it was protecting you from.

• Upgrade it so the protection stays but the friction drops.

• Confirm with a small reversible test.

Then decide:

Keep. Upgrade. Remove.

Less impulse.
Less regret.
More freedom that actually holds.

• The Lindy Effect is a filter for non-perishable things: ideas, books, principles, skills, and habits.

• It says: if something has survived for a long time, it is more likely to keep surviving.

• Lindy predicts persistence, not quality. Old is not automatically best.

• The practical move is core vs wrapper: keep the durable core, hold the wrapper loosely.

• AI helps you do this with a simple loop: extract → plan → confirm.

After reading, you will have a simple filter plus three copy-paste prompts to decide in minutes: ignore, small test, or commit.

1) Learning framing: the goal is calmer choices in a fast-changing world (with AI)

A new story hits your feed.

In early February 2026, a “social network for AI agents” called Moltbook went viral. Screenshots spread. People reacted like it was a glimpse of the future. Excitement, fear, memes, hot takes. The emotional tempo was the point.

Then the spell weakened.

It became clear how easy this kind of “agent society” vibe is to manufacture. Some of the most viral posts were human-directed. The platform also filled up with familiar internet behavior: spam, scams, and attention grabs.

That leads to a simple question:

What makes online things durable, not just viral?

If you look at what people actually come together for, the core is stable: trust, shared context, belonging, reputation, real connection. When that core is missing, the wrapper takes over, and the wrapper is easy to game.

Hype is loud. Time is honest.

That is why the better question is this:

What part of this will still matter in 5 years?

That question is the heart of this week’s mental model: the Lindy Effect.

How AI helps

AI helps you think in a cleaner sequence:

• Extract the durable core from the noisy wrapper.

• Plan a small, low-regret test instead of committing inside a hype cycle.

• Confirm with evidence after time passes, so you keep what truly works and drop the rest.

2) The Lindy Effect in plain language (and why it works)

The Lindy Effect, also called Lindy’s Law, is a simple idea.

For non-perishable things, time is evidence. If something has survived for a long time, it is more likely to keep surviving.

Speed bump, because it prevents the most common misunderstanding:

Lindy predicts persistence, not quality.

A helpful rule of thumb is:

If a non-perishable has lasted X years, expect roughly X more.

The key is the phrase non-perishable. This model is not about physical objects that wear out. It is about things that can keep being reused without “spoiling.”

Three quick tests help:

• Can it be copied or reused without degrading?

• Is it mostly information or behavior, not a consumable?

• Does it survive interface churn?

This is why old books that people still recommend are often worth your attention. They have been tested by changing tastes, criticism, and competition. A brand-new trend has not been tested yet.

One more guardrail keeps you honest. Longevity can come from lock-in, habit, and switching costs. Treat Lindy as a ranking input, then use real evidence when you have it.

The practical payoff is simple.

When you feel overwhelmed by novelty, Lindy gives you a calmer default. It pulls your attention toward what has earned its place over time.

And in fast-changing spaces, it gives you a smarter move: apply Lindy to the principle beneath the tool, not the tool of the week.

3) Prompts: use AI to extract the durable core, make a plan, and validate it (extract → act → confirm)

The Lindy Effect becomes useful when you can apply it quickly.

That is what AI is great for. It can help you separate the durable core from the noisy wrapper in minutes, and then turn that into a small experiment you can actually run.

Do not run all prompts at once. Pick the stage you are in:

• Prompt 1 = understand

• Prompt 2 = act

• Prompt 3 = decide

3.1 Diagnostic prompt: extract the non-perishable core (analysis)

Use this when you see something new and want clarity fast.

Apply the Lindy Effect to this thing.

Return answers in three buckets:
A) Known (explicitly supported by what I pasted)
B) Inferred (reasonable, but not stated)
C) Guess (speculation — label it)

1) Describe it in one sentence.
2) Identify the non-perishable core (what would still matter in 5–10 years).
3) Identify the perishable wrapper (what is likely to change, get gamed, or vanish).
4) List 3 older analogs to the core (or say “no clear analogs yet”).
5) Recommendation: ignore / small test / invest attention — and why.
6) Bet size + confidence:
   - Bet size: small / medium / large
   - Confidence: low / medium / high

CONTEXT (paste article, notes, screenshots, claims):
[PASTE HERE]

3.2 Build prompt: turn the core into a 7–14 day plan (action)

Use this when the core looks real and you want to try it without getting pulled into the wrapper.

Turn the Lindy “core” into a small plan I can actually run.

1) State the non-perishable core in one sentence.
2) Write the core as a measurable outcome (how I’ll know it helped).
3) Propose ONE durable default (a rule or habit) that captures the core.
4) Give me a 7–14 day experiment:
   - Day 1 setup (under 30 minutes)
   - Daily checklist (max 3 items)
   - One measurement
5) Tell me what to ignore for now (the perishable wrapper).
6) Add one friction fix (how to make it easier in real life).

CONTEXT & MY CONSTRAINTS (time/energy/budget):
[PASTE HERE]

3.3 Lindy follow-up prompt: confirm durability after time passes (decision)

Use this after you tried the plan, or after the hype has had time to cool down.

Help me decide if this deserves to become a durable default (Lindy follow-up).

1) Restate the non-perishable core and the perishable wrapper as they look now.
2) Summarize the evidence I gathered (not vibes).
3) Decide: Keep / Iterate / Drop — and why.
4) If Keep: propose a “maintenance version” I can do long-term with low effort.
5) If Iterate: change ONE variable and give me the next 7-day test.
6) If Drop: extract the lesson and suggest a better Lindy-leaning alternative.

WHAT I TESTED & RESULTS / NOTES (paste):
[PASTE HERE]

4) Principles and traps

The Lindy Effect works best as a calm filter, then a small experiment.

Principles

• Rank, do not prove.
  Lindy helps you decide what deserves attention when certainty is impossible. It does not settle the truth.

• Core vs wrapper.
  The core compounds. The wrapper rotates. When something new appears, ask what stays valuable if the product disappears.

• Match bet size to evidence.
  New and untested means a small bet. Old and repeatedly useful earns a bigger bet.

• Barbell your attention.
  Keep most of your life on durable foundations, and reserve a small budget for experiments.

• Wait for the chorus.
  Do not reorganize your life around one loud week. Let time reveal what keeps showing up.

Traps

• Survivorship bias.
  You see the winners. You do not see the graveyard. Do not treat survival as the only signal.

• Lock-in.
  Some things persist because leaving is painful, not because they are great. Longevity can predict stickiness without implying quality.

• The anti-new reflex.
  Lindy is not a reason to dismiss new things. It is a reason to size your bets.

• Wrong unit.
  Apply Lindy to the principle, not the interface. Many “new tools” are old principles in new packaging.

• Staying abstract.
  If you never run a small test, you never learn. Turn insights into a plan, then decide based on evidence.

5) Closing: the practical payoff

You are not behind. The surface is loud.

When new things are easy to create, it gets easier to manufacture excitement, proof, and popularity. That does not mean you should become cynical. It means you need a calmer filter.

The Lindy Effect gives you that filter. It helps you invest attention in things that have earned its place over time, and it helps you treat novelty with the right bet size.

Use the loop:

• Extract the durable core.

• Act with a small plan.

• Confirm with evidence.

• Lock it in as a durable default.

Hype is loud. Time is honest.

That is the payoff. Less whiplash. More compounding.

Share

TL;DR

• Defaults are not neutral. They quietly become the “baseline” your brain protects.

• The Reversal Test is one move: if a change in one direction seems bad, test the opposite.

• If both directions seem bad, you are often tuning the wrong dial.

• AI helps you do this fast with a simple loop: analyze, propose, pressure-test.

1) Learning framing: the goal is cleaner everyday decisions (with AI)

You try to fix something simple, and you get stuck.

Move the deadline earlier and everyone stresses. Move it later and nothing happens until the last minute.
Spend less and you feel deprived. Spend more and you feel reckless.
Have fewer meetings and people feel disconnected. Have more and everyone gets exhausted.

The weird part is not the details. It is the pattern.

Both directions feel wrong.

This guide gives you a fast way out of that loop. In a few minutes, you can spot the hidden default underneath the decision, test both directions, and design a better setup you can actually try.

Defaults are not neutral. They are just familiar.

A default is any setting, rule, or habit that became “normal” over time, even if nobody chose it on purpose. Once it becomes the baseline, your brain starts treating it like the sensible middle.

This week’s mental model is the Reversal Test, and it is deliberately small.

How AI helps

AI is useful here for two jobs:

• Make the default visible and clarify what you are trying to achieve.

• Suggest a better setup so the right behavior becomes easier.

2) The Reversal Test in plain language and why defaults trap you

The Reversal Test is simple.

If changing something in one direction seems bad, test the opposite direction.

That alone helps because it interrupts the feeling that the current setting is obviously correct.

The real value shows up when both directions seem bad.

If both directions seem bad, the current setting is being treated as “just right” without a clear reason. That is a strong sign you might be tuning the wrong dial.

The basic move

Pick the dial you are trying to tune:

• earlier vs later

• more vs less

• stricter vs looser

• faster vs slower

Now run the reversal:

1. Imagine the change you were considering. Why does it seem bad?

2. Imagine the opposite change. Why does that seem bad?

If only one direction is bad, you have a clearer direction.

If both directions are bad, you have a signal. It usually points to one of these:

• The goal is unclear.

• A real constraint is not being said out loud.

• The dial is a proxy for something else.

• The setup needs a redesign.

Your quick example captures it well.

If moving a deadline earlier is bad and moving it later is bad, the issue may be the deadline system itself. A single date is doing too much work.

Two everyday examples

1) Budgeting
Spending less feels miserable. Spending more feels reckless.
A clean redesign is to remove daily decision pressure with one default: set an automatic transfer on payday into savings and bills, then treat what remains as guilt-free spending.

2) Screen time
Stricter rules fail. Looser rules drift.
A clean redesign is to change the environment: turn off non-essential notifications and keep distracting apps off your home screen, so “less screen time” happens by default.

This is the practical benefit of the model.

You stop looping on direction and you start creating setups that make the right behavior easier.

3) Prompts: use AI to spot default traps and redesign the setup

The simplest way to use this with AI is one sentence.

Paste what is going on, then add:
Apply the Reversal Test. Help me identify the default I am protecting, test both directions, and tell me if I am tuning the wrong dial.

Below are three short prompts that do the same loop every time: spot, rebuild, pressure-test.

3.1 Diagnostic prompt: apply the Reversal Test to what I wrote

Use this when you feel stuck between two directions and both feel wrong.

Apply the Reversal Test to my situation.

1. Identify the default I am treating as normal.

2. State my goal in one clear sentence.

3. List the real constraints you infer.

4. Test the change in one direction, then the opposite direction.

5. If both directions seem bad, tell me what dial might be wrong and what I should reconsider.

Keep it plain language. Do not force symmetry.

[SITUATION]

3.2 Build prompt: redesign the setup so the right behavior is easier

Use this when both directions feel bad, or when the same problem keeps repeating.

Help me redesign this situation so I stop tuning the same dial.

Give me:

1. One new default or rule I can adopt.

2. One small environment change that makes it easier to follow.

3. A 7-day experiment with one simple measurement.

4. A safety plan so I can revert if it backfires.

Keep it realistic and low effort.

[SITUATION]

3.3 Reversal Gauntlet prompt: pressure-test the redesign before I commit

Use this after you have a proposed new default and want to make it sturdier.

Pressure-test my proposed new default like a practical, friendly critic.

1. List the top 5 ways this fails in real life.

2. Point out any real asymmetries where reversal thinking does not apply.

3. Suggest one simplification that keeps the benefit but reduces friction.

4. Tell me what evidence would justify keeping it after 7 days.

[PROPOSAL]

4) Principles and traps

The Reversal Test works best when you treat it as a fast way to clarify what is going on, then act with a small experiment.

Principles

• Reverse in realistic steps.
  Test the opposite direction in a way you could actually choose.

• Say the goal out loud.
  If you cannot name what “better” means, every direction will feel wrong.

• Name constraints early.
  Time, money, energy, relationships, and rules explain most “this won’t work” reactions.

• Treat “both bad” as a clue.
  It often points to the wrong dial. The fastest win is redesigning the setup.

• Prefer new defaults over willpower.
  A good setup makes the right choice easier without constant self-control.

Traps

• Forcing symmetry.
  Sometimes one direction really is worse.

• Using it to win an argument.
  The point is shared clarity and better choices.

• Treating “both bad” as “do nothing.”
  It usually means you need a different setup and a smaller first step.

• Ignoring hidden costs.
  Switching costs and social friction can be real. Make them explicit.

• Going too abstract.
  Keep it grounded in one situation. Find one new default you can actually try.

5) Closing: the practical payoff

Heavy decisions often stay heavy because they have no structure.

You see two directions, both feel wrong, and you postpone. Not because you are lazy, but because you do not know what you are optimizing, what you are protecting, or what breaks if you commit.

The Reversal Test with AI fixes that fast:

• Analyze: surface the default, the real goal, and the constraints.

• Propose: if both directions feel bad, redesign the setup with a better default.

• Pressure-test: find what could go wrong while it is still cheap to adjust.

I ran this on a decision I had been carrying for too long: consultancy versus startup. The prompts did not decide for me. They gave me clarity and a safe path to act.

That is the point.

Not perfect answers. Deeper thinking, slower and faster, when it matters.

Last week I watched an AI answer turn into a confident plan, then collapse on the first edge case. It looked finished. It was not.

Dunning–Kruger is a lag between doing and judging. Your performance can improve faster than your ability to evaluate your performance. Calibration closes that gap.

This guide shows how to use AI to go deeper faster while staying calibrated:

• Build a Depth Map so you stop learning randomly.

• Convert reading into micro-tests that expose misconceptions early.

• Use AI as a Depth Stress Test to see if your confidence is earned.

• Add one Reality Anchor each week so you do not confuse polished output with real skill.

1) Learning framing: the goal is earned confidence with AI

AI is the best confidence machine ever built because it is fluent.

When you are new to something, fluent explanations are hard to distinguish from expertise until reality collects its fee.

This week’s mental model is Dunning–Kruger. I do not want the pop version. No stereotyping. No dunking on beginners.

The useful version is practical.

When you start learning a topic, you can improve your output quickly while your ability to judge that output improves more slowly. That gap is where overconfidence lives.

AI can make the gap feel smaller than it is, because it can explain things cleanly, generate plausible examples, and produce answers that look finished. That is power if you use it correctly.

Here is the frame for the rest of the guide.

AI has two roles

1. Generator: It helps you move fast: summaries, drafts, plans, solution attempts.

2. Examiner: It helps you find where you are bluffing: tests, edge cases, transfer questions, “prove it” tasks.

Most people mostly use Generator mode. They learn faster and they get confidently wrong faster.

Your advantage is simple. Use AI to move quickly and to test yourself continuously.

The rule that keeps this honest

AI is a sparring partner, not a referee.

You use it to stress-test your understanding, then you anchor what you learned in reality with one external check:

• a real task outcome

• a trusted reference

• or a human who knows the domain

Next we will translate Dunning–Kruger into plain language and make it usable. You will learn to separate “I can repeat it” from “I can use it.”

2) Dunning–Kruger in plain language and why it shows up when learning with AI

Dunning–Kruger is simple.

When you are new to a topic, you do not just lack skill.
You also lack the internal meter that tells you how much skill you lack.

That is why early confidence is often cheap. You can follow an explanation and your brain tags it as understanding.

The mechanism: judging lags doing

In many domains, the skills that let you do the work are the same skills that let you evaluate the work.

So beginners can produce something that looks right and still miss the errors. The error detectors have not formed yet.

What AI changes

AI compresses the feeling of progress.

You get clean explanations, plausible examples, and answers that look finished. That can be great for speed.

The risk is specific. You can generate fluent output before you have built the judgment to audit it.

So the move is not to slow down. The move is to learn in a way that forces calibration.

A quick test for real depth

If you want to know whether you actually understand something, try one:

• Give one example and one non-example.

• Predict what happens if one condition changes.

• Handle an edge case.

• Apply it in a new context without looking.

If you stall, that is signal. It shows you where to aim next.

Now we turn this into a workflow you can run quickly.

3) Prompts: use AI as a learning accelerator and a calibration tool

The fastest way to learn with AI is to use it in two roles:

• Generator to move fast

• Examiner to reveal what you do not yet see

Here are three prompts. Run them in order.

3.1 Diagnostic prompt: build a Depth Map (stop random-walk learning)

Use this when you start a new topic and want a clear path.

Copy and paste:

Build me a Depth Map for this topic.
Define beginner, intermediate, and advanced as concrete capabilities.
For each level, list:

1. the core concepts

2. the top 5 misconceptions

3. 3 proof tasks that demonstrate real competence

4. the prerequisites I should not skip

Keep it plain language and practical.
End with the smallest next step I can do in 30 minutes.

Topic: [PASTE TOPIC HERE]

What you want is a map where every level has proof tasks. Proof tasks keep confidence attached to evidence.

3.2 Build prompt: turn the topic into micro-tests (learn by retrieval)

Use this after you read or watched something and want fast feedback.

Copy and paste:

Generate micro-tests for this topic at my current level.
Create 10 questions that reveal misconceptions.
Include:

• 3 questions that look easy but are traps

• 2 edge cases

• 2 transfer questions (apply in a new context)

For each question, provide:

1. the correct answer

2. why the common wrong answer is tempting

3. what concept the question is testing
   Keep them short.

Topic: [PASTE TOPIC HERE]
Level: [BEGINNER / INTERMEDIATE]
What I just learned: [PASTE NOTES OR A LINK SUMMARY]

Run these like reps. The goal is not reassurance. The goal is signal.

3.3 Depth Stress Test prompt: interview me like a senior (AI as sparring partner)

Use this to check whether your confidence is earned.

Rule: you answer first. AI grades after.

Copy and paste:

Run a Depth Stress Test on me for this topic.
Ask 8 questions with increasing difficulty.
Include 2 edge cases and 2 transfer questions.
After each question, wait for my answer.
Only after my answer, grade it using this rubric:

• assumptions stated

• correctness

• handling of edge cases

• ability to transfer

• ability to predict what changes under a modified condition

After all 8 questions, tell me:

1. what I genuinely understand

2. where I am likely overconfident

3. the single highest-leverage misconception to fix next

4. one proof task I can do this week to validate in reality

Topic: [PASTE TOPIC HERE]
My current understanding in 5 bullets: [PASTE HERE]

Share

Reality Anchor

Once per week, attach your learning to something external: one proof task, one trusted reference check, or one knowledgeable human review.

Next section, we will add principles and traps so you get the upside without drifting into fluency-as-mastery.

4) Principles and traps

The prompts work when you follow a few tight rules. Think of this as the guardrail that keeps learning fast and honest.

Principles

• End every session with calibration
  One micro-test, one prediction, or one proof task. Do not stop at “nice explanation.”

• Prefer proof tasks over explanations
  Proof tasks create inspectable outcomes: a working script, a fixed bug, a solved set, a decision that survives reality.

• Demand predictions
  Ask “What changes if I change X?” If you cannot predict, you do not own it yet.

• Stress beats rereading
  Rereading creates fluency. Stress testing creates signal.

• Match rigor to stakes
  Low stakes: Depth Stress Test plus micro-tests.
  High stakes: add redundancy with docs, trusted sources, or a human check.

• One Reality Anchor per week
  If learning never touches reality, it stays in story mode.

Traps

• Using AI for reassurance
  “Does this look right?” is comfort. Ask it to break your understanding.

• Confusing clean language with depth
  Depth shows up in edge cases, constraints, and transfer.

• Endless mapping
  Map once, then test. If you map twice, you ship once.

• Letting AI be the referee
  Use it as a sparring partner. Anchor on outcomes and trusted references.

Finally, we compress it into a weekly loop so it compounds.

5) The weekly practice: a 10-minute Depth Check that compounds

You do not need a new system every week. You need one small loop that keeps calibration alive.

Once a week, set a 10-minute timer and run this Depth Check.

The five questions

1. What did I think I understood this week?

2. What did I prove I can do?

3. Where did I hesitate, hand-wave, or get surprised?

4. What is one misconception or weak spot to fix next?

5. What is one proof task I will do next week to anchor this in reality?

That is the whole model. The goal is to keep confidence attached to evidence.

This week

Pick one topic you are learning right now.

• Build the Depth Map once.

• Run one micro-test set.

• Do one proof task.

• End the week with the Depth Check.

That is how you learn faster with AI while keeping your confidence earned.

TL;DR
If you’ve ever seen one signal (a scary headline, one bad review, one symptom) and instantly felt “this changes everything,” this is for you.

The mistake is simple: one vivid signal hijacks your judgment.
The fix is simple: start with the baseline (what usually happens), then decide what to do.

This is a small upgrade you can use every day:

• a quick check that makes your beliefs calmer and more accurate

• a way to take small, low-regret actions without spiraling

• a few simple AI prompts to think deeper, not faster

Framing: stop letting one signal steer the whole story

Most overreactions have the same shape.

Something happens.

Your brain grabs it.

A story appears fully formed: this means X… so I should do Y… right now.

Two normal-life versions:

1) The headline spike

You see: “DOG LANDS PLANE AFTER PILOT FAINTS — Experts Shocked.”

For 30 seconds, your brain goes: *Wait, are we living in a different world now?*

Then the quieter truth returns: most days, life stays mostly the same. Outliers just get microphones.

2) The one scary health signal

You google a symptom and get: “THIS COULD BE A SIGN OF A SERIOUS CONDITION.”

Instantly your mind jumps to the worst story.

But one signal isn’t a verdict. Rare things are still rare, and false alarms happen.

The goal isn’t to be numb. It’s to add one small pause that asks a better first question:

Before I react to this signal… what usually happens in situations like this?

That question is the whole upgrade. Next, I’ll explain it in plain language and show how to use it (with AI).

Base rates in plain language (with tiny examples)

A base rate is the baseline: how often something is true in real life.

A signal is the thing you just saw: a headline, a message, a symptom, a test result.

Base rate neglect is what happens when you let the signal do all the talking.

Here’s the simple upgrade:

Start with the baseline. Then let the signal adjust you.

Example 1: the headline spike

Signal: “DOG LANDS PLANE — Experts Shocked.”
Baseline: most headlines are outliers, and most days your actual life doesn’t change.
Better take: “Interesting. Probably not a new era.”
Low-regret action (if it matters): check again tomorrow, or look for a second reliable source before you change any belief or plan.

Example 2: the scary health signal

Signal: “This could be serious.”
Baseline: most symptoms have common explanations; false alarms happen.
Better take: “I shouldn’t ignore it, but I also shouldn’t jump to the worst story.”
Low-regret action: do one calm next step (write symptoms down, check for clear red flags, or book a professional check if needed).

One more important detail: baselines don’t have to be precise.

If you don’t know the exact base rate, you can still use a rough range:

• rare

• sometimes

• common

That alone stops the signal from hijacking your brain.

Next, we’ll make this practical: a few simple prompts to use AI as a thinking partner (pause → baseline → decide).

Prompts: use AI as a thinking partner (pause → baseline → decide)

AI is useful here for one reason: it can slow you down and show you alternatives when your brain is locked onto one story. You are not asking it for the truth. You are asking it for a clearer way to think.

3.1 Simple prompt: “What’s the baseline?”

Use this when something feels urgent and you notice your mind spinning a story.

I might be overreacting. Help me sanity-check this.
What’s the usual outcome in situations like this?
What’s a more realistic take than my first reaction?
What’s one small thing I should do before I decide?

Situation: [what happened + what you’re about to do]

3.2 Simple prompt: “How common is this (roughly)?”

Use this when you do not need precision. You just want to stop guessing wildly.

Help me estimate how common this is, in plain language.
What group should I compare to (people like me, in this situation)?
Give me a rough range (rare / sometimes / common) and why.
Then tell me what I should do if it’s rare vs common.

Situation: [signal + decision]

3.3 Inverted prompt (compact): “Beat the base rate”

Use this when you do not just want the most likely outcome. You want to do better than default.

Help me beat the base rate here.

1. If I do what most people do, what’s the likely outcome?

2. What do most people do that keeps them stuck there?

3. Give me 3 specific behaviors to outperform the baseline (one can be outside-the-box).

4. For each: the trade (cost/risk) + the upside.

5. Pick one for me and turn it into a 7-day plan (what I’ll do + when + what to track).

Situation: [1–2 sentences + constraints]

3.4 Action prompt (compact): “Make a calm plan I’ll actually follow”

Use this when you want a response that improves your life even if your scary story is wrong.

Help me respond in a calm, useful way.

1. What’s the most likely normal outcome here?

2. What’s one small action I can take today (10 minutes) that helps either way?

3. What’s one simple rule to prevent overreacting next time?

4. What should I check in 7 days to see if this worked?

Situation: [one sentence about what happened + what I feel like doing]

4) Principles and traps

Principles

• Start with “what usually happens”

  Before you react to a signal, name the baseline. Your first story is rarely the most accurate one.

• Default to baseline for beliefs, prepare for tails with actions

  Most of the time, the boring outcome wins. If the stakes are high, take a small, low-regret step while you learn more.

• Use the right comparison group

  Not “people in general.” People in your situation. Same context, same constraints.

• If you do not know the baseline, use a rough range

  Rare, sometimes, common. That is enough to stop wild overreactions.

• Get one more datapoint before you commit

  Another day, another source, another opinion, another check. One extra data point beats one dramatic signal.

Traps

• One vivid story overrides reality

  Outliers get microphones. Your brain treats them like the new normal.

• The wrong baseline

  You used a reference that does not match your situation, so the “calm conclusion” is fake calm.

• Old baselines

  Sometimes the environment really changes. If the context shifted, treat the old baseline as a starting point, not a rule.

• All or nothing thinking

  “This changes everything” or “this means nothing.” Real life is usually in the middle.

• Using the baseline as an excuse to do nothing

  Baselines are for better decisions, not for passivity. If something matters, take the small step and collect better signal.

The weekly practice that keeps you grounded

Once a week, take 10 minutes and do a quick reset.

Ask yourself:

• What signal did I overreact to this week?

• What was the baseline I ignored?

• Where did “usually nothing changes” turn out to be right?

• Where might the baseline be shifting?

• What is one small check I will do next time before I decide?

That is it.

If you do this weekly, you start to notice a pattern: most of your stress comes from treating a single signal like a verdict.

This week: pick one area (news, health worry, work decision). The next time something spikes your emotions, do the baseline check once, then take one small low-regret step.

TL;DR
If you have tried New Year goals and bounced by week 3, this is for you. This is not motivation. It is a system that survives low-energy days.

Your life already runs on loops. This year gets easier when you build loops that compound and install guardrails that catch drift early.

• Reinforcing loops amplify what is happening.

  Helpful: 10-minute walk → more energy → you walk again.

  Harmful: scroll late → worse sleep → you scroll more.

• Balancing loops stabilize what is happening and pull you back toward a default.

  Helpful: “If I want to buy it, I wait 24 hours” → impulse drops → spending stabilizes.

  Harmful: feel discomfort → avoid the task → relief → avoidance becomes the default.

• Delays are why change feels fake at first. Most people quit in the “nothing yet” window.

New Year framing: build a year that helps itself

The real upgrade is a setup where tomorrow becomes easier than today.

Here is what that looks like in normal life.

You start walking for ten minutes after lunch. Nothing heroic. After a few days you notice you sleep a bit better, your mood lifts, and the next walk costs less effort to start. It stops being a decision and starts being “what you do.”

That is the shape of a good year. Small effort creates a small benefit, and that benefit makes the next repetition more likely.

This mental model is just a way to spot those loops on purpose, then edit them.

Feedback loops in plain language (with tiny examples)

A feedback loop is simple:

What happens today changes the odds of what happens tomorrow.

You will sometimes hear the terms “positive feedback” and “negative feedback.” Ignore the vibe of those words. They mean “amplifying” and “stabilizing,” not “good” and “bad.” In this guide, we will call them reinforcing and balancing, and we will label them separately as helpful or harmful.

Reinforcing loops (they amplify)

These loops make the next step more likely. They compound in one direction.

• Helpful reinforcing loop:

  10-minute walk → more energy → you walk again → fitness improves → walking feels easier.

• Harmful reinforcing loop:

  scroll late → worse sleep → lower energy → you reach for the easiest reward → you scroll more.

Balancing loops (they stabilize)

These loops pull you back toward a default. Think of a default as the level you drift back to.

• Helpful balancing loop:

  “If I want to buy it, I wait 24 hours” → impulse drops → spending stabilizes.

• Harmful balancing loop:

  feel discomfort starting a task → avoid it → instant relief → avoidance becomes the default.

Delays (why change feels fake at first)

Many loops have a delay between action and result. You do the walk today and you do not feel fitter tomorrow. You eat better for a week and the scale barely moves. You practice a hobby and still feel clumsy.

Decide a minimum evaluation window upfront (7, 14, or 30 days) and track one leading indicator you can notice sooner (sleep quality, morning energy, cravings, consistency). Otherwise you quit in the delay window.

It is 11:48pm, you tell yourself “one more video,” and tomorrow already feels heavier. Put the charger outside the bedroom and replace the last five minutes with something boring on purpose, and the loop starts to break.

Next, we will use AI to do the practical part: map your loop, install one compounding loop, and break one draining loop by changing a single link.

Prompts: use AI as a loop designer (see it → build it → break it)

One warning: a loop map is a hypothesis, not proof. AI can confidently invent clean stories, so you treat its output as a starting point and validate it with small tests.

Diagnostic prompt: “What loop am I in?”

Use this when you feel stuck in a pattern and you cannot tell what is driving it.

Map my situation as feedback loops.
Identify (1) the reinforcing loop that is helping, (2) the reinforcing loop that is hurting, and (3) one balancing loop I could add as a guardrail.
Keep it simple and name the links in plain language.
For each link, list the assumption you are making and what evidence would confirm or deny it.

Situation: [sleep, fitness, diet, hobby, work, mood + what has been happening]

The goal is one clear sentence you can test, like: “Late scrolling hurts sleep, and bad sleep makes me reach for late scrolling.”

3.2 Build prompt: “Install a tiny reinforcing loop that compounds”

Use this when you want progress that does not require willpower every day.

I want a minimal-effort habit that compounds over 12 months.
Propose a 5-minute daily version, an immediate reward, and one change that makes tomorrow easier than today (environment, prep, identity).
Also give me a bad day version (2 minutes) so the loop never breaks.
Define a 14-day evaluation window and one leading indicator I can notice early.

Goal: [walking, learning, hobby, health]

This prompt forces the most important design rule: the loop has to survive bad days.

3.3 Break prompt: “Cut one link, do not rely on willpower”

Use this when a pattern keeps repeating and you are tired of trying to “want it more.”

Help me break this draining loop by changing just one link.

1. Restate the loop in one sentence.

2. Show 3 places to intervene:

   • add friction (make the bad input harder)

   • replace with a smaller action (a 2-minute alternative)

   • add a guardrail rule (a balancing loop)

3. Pick the best intervention for minimum effort and maximum impact, and tell me exactly how to implement it this week.

4. Tell me what real-life constraints might get in the way, and how to adjust without giving up.

Loop: [describe the pattern]”

You are not trying to fix your personality. You are editing one link in the chain.

Principles and traps

Principles

• Make the next rep easier than the last

  Shoes by the door. Guitar on a stand. Healthy food visible. Phone out of the bedroom.

• Use guardrails, not punishment

  Calm defaults beat drama: a 24-hour buying rule, a screen cutoff, a two-minute version on tired days.

• Respect delays

  Pick an evaluation window and track one leading indicator, not just the final outcome.

• Change one link at a time

  One tweak per week gives you signal.

• Scale down before you quit

  If it creates strain faster than capacity, reduce the dose until it is sustainable.

Traps

• Reinforcing does not mean helpful

  Doomscrolling compounds too. Always label loops as helpful or harmful.

• All or nothing resets

  Perfect plans create rebounds. Small corrections win.

• Treating AI as truth

  Use AI to generate hypotheses. Validate with your lived data.

• Mapping forever

  A loop map is only useful if it produces one small experiment this week.

The weekly practice that keeps the year moving

A small weekly check-in keeps you steering.

Once a week, take ten minutes and ask:

• What loop helped me this week?

• What loop hurt me this week?

• What is one link I will change next week?

• What is my bad day version so the loop survives?

• What evaluation window am I using, and what is my leading indicator?

If you want to use AI for the review:

Here is how my week went: [short bullets].
Identify one helpful loop and one harmful loop.
Suggest one link to change next week.
Keep it small enough that I will do it even on a bad day.

That is the whole model.

This week: pick one arena (sleep, diet, fitness, hobby). Map the loop once. Make one change for seven days. Let the loop do the heavy lifting.

TL;DR

Problem: Most of us slowly overbuild our lives and projects: more routines, more apps, more offers, more “good ideas”, until even useful things turn into background stress. AI often makes this worse. Ask for a better morning routine or a plan to grow your small yoga business and it will happily give you something detailed, ambitious, and completely unrealistic for an ordinary week.

What this gives you: Via negativa is a simple mental model. Instead of asking “What else could I add?”, you start from what you already have and ask “What can I remove so this still works, but feels lighter?” This guide shows you how to simplify your life with AI by turning it into a subtraction engine: something that lays out your current week, routines, or ideas, highlights what actually matters, and suggests what you can safely remove, shrink, or postpone. By the end, you will have three copy-paste prompts you can drop into any AI chat to make one part of your life or work lighter in under 10 minutes, plus a tiny habit for using them regularly.

1. The Slow Creep of “Too Much”

There is a particular kind of tired that comes from having slightly too many good ideas in your life.

You do not crash. You do not burn out.

Subscribe now

You just… drag a little.

Your morning routine used to be “wake up, stretch for five minutes, coffee”.

Then you saw a thread about the perfect morning and added journaling.

A podcast convinced you to add breathwork.

Someone mentioned cold showers.

AI helped you build a 12 step “optimal start to the day” with affirmations, mobility, planning, and three different apps.

None of these are stupid. Each one has a reason.

But now your mornings feel like you are clocking in for a shift instead of just starting your day.

The same thing happens with small projects and businesses.

You have a simple idea:

“I want to teach yoga and earn a bit of money from it.”

You open an AI assistant and ask for help.

It eagerly gives you:

• A brand concept and colour palette

• A content calendar for three social platforms

• A website structure with blog, FAQ, and testimonials

• Four pricing tiers

• A plan for email marketing and a referral program

Again, nothing is wrong with any single piece.

But by the end of the conversation, “teach a few local yoga classes” has turned into “launch a fully fledged wellness brand”.

You still have not taught a single class.

You definitely have not made your first dollar.

But now you have a backlog of things you feel guilty for not doing.

One reasonable addition at a time, life drifts from simple and doable to busy and slightly unlivable. This guide is about using via negativa, together with AI, to quietly reverse that drift: to map what you already have, see what actually matters, and get concrete suggestions for what you can remove, shrink, or postpone without losing the result. In the next section, we will make via negativa itself very simple, then turn it into practical prompts you can run on your own life and small projects.

2. Via negativa in plain language

The classic idea of via negativa is:

Improve something by taking things away, not by adding more.

In everyday terms:

Start from the life or project you already have and ask:

“What can I remove so this still works, but feels lighter and easier?”

You are not trying to become a monk or win minimalist points.

You are trying to get closer to the minimum version that still does the job:

• The simplest morning that still leaves you feeling okay about the day.

• The simplest version of your yoga business that still pays you.

• The simplest set of commitments that still makes your week feel like your week.

If you feel behind all the time, it is usually not a personal failure.

It is a sign the system around you is carrying more than it needs to.

Most of us flip this.

We ask “What else could I add?”

New apps, new exercises, new offers, new platforms.

Via negativa is the habit of pausing and asking first:

“Before I add anything, what could I quietly remove?”

AI fits into this in a very specific way.

It is extremely good at laying everything out on the table:

• All the things you do in a week

• All the steps in your routine

• All the ideas you have for a small business

Once it has the map, you can ask it to be blunt:

• Which parts look optional?

• Which parts are doing the same job twice?

• Which parts are heavy compared to the value they bring?

The model does not tell you how to live.

It just surfaces the cuts you could make, so you can decide.

When it helps

Use via negativa when:

• Your day feels full but strangely unsatisfying.

• You have a plan from AI that looks great on paper and impossible in real life.

• You have a simple offer, like classes, lessons, sessions, buried under too many “nice to have” extras.

• You keep saying “I do not have time”, but you never actually list what is filling the time.

In those cases, via negativa is a first pass with the eraser.

It does not redesign your whole life. It just asks, “What can we remove right now without breaking the point?”

When to be careful

Be careful with this lens when:

• You are tempted to delete the one thing that genuinely helps because it is slightly effortful.

• You are using “simplify” as a way to avoid doing anything at all.

• The real problem is “I do not know what I want”, not “I have too much”.

• You are in a situation where the direction itself is wrong and needs rethinking, not pruning.

Via negativa is not about refusing to grow or never adding anything new.

It just asks you to remove the unnecessary weight first, so whatever you do add has room to breathe.

Next, we will turn this into concrete AI prompts you can run on your own week, your routines, and simple “first dollar” business ideas.

3. Prompts to use via negativa with AI

The easiest way to use this mental model with AI is to literally say:

“Apply via negativa and help me see what I can remove.”

You describe a part of your life or work, add that one line, and let the assistant map things out and suggest what could go. Under the hood, all of these prompts follow the same pattern: List → Label → Lighten → Test. You list what is there, label what each thing is for, lighten the load by cutting or shrinking something, then test it for a week.

These are copy paste prompts. Save or screenshot them so you can reach for them the next time you want to simplify your life with AI, because your week, your routines, or your small business idea starts to feel heavier than it should.

3.1 Quick subtraction pass on your week

Use this when your week is full, you are always “busy”, but you are not sure what is actually worth it.

Apply via negativa to my week.

1. List my main recurring activities and what each is meant to give me (for example income, connection, learning, rest).

2. Mark the 3 to 5 that look least essential or could be merged with others.

3. Suggest 2 to 3 experiments where I remove, reduce, or bundle something for one week and tell me what to watch for to see if I actually miss it.

Here is roughly what a normal week looks like for me:
[describe your typical week: work hours, recurring calls, classes, chores, social things, side projects]

One week in, one lighter version of that week out.

If you try cutting something and nothing important breaks, it probably did not need to be there in the first place.

3.2 Trimming an overcomplicated routine or plan

Use this when you asked AI, or the internet, for a “better routine” and got something that looks great but you already know you will not follow on a tired day.

Apply via negativa to this routine.

1. Restate what this routine is trying to achieve in plain language.

2. Mark the few steps that probably create most of the benefit.

3. Build a ‘bare minimum’ version I could realistically do even on a busy, low energy day.

4. Put the remaining steps into an optional ‘good day’ version, not mandatory.

Here is the routine or plan:
[paste or describe your current morning routine, workout plan, study schedule, etc.]

You are not asking AI for a more impressive routine.

You are asking for the smallest version that still helps, plus extras you can add when you actually have the energy.

3.3 Turning a big small business idea into “first dollar this week”

Use this when you have a simple, human size idea, like:

• Teaching yoga or pilates

• Offering language or music lessons

• Selling baked goods from your kitchen

• Doing simple freelance work

and it has exploded into a full “brand” in your head.

Apply via negativa to this small business idea.

1. Summarize in one or two sentences what real value I want to give people.

2. Design the simplest possible offer I could realistically sell in the next 7 days (for example, one kind of class, one type of session, one product).

3. List everything else that can be safely postponed until after I have made my first dollar, and explain what I gain by not doing those things yet.

Here is my idea and everything I think I need to do:
[describe your idea, all the offers, platforms, branding tasks, content plans, etc.]

Instead of “launch a full wellness brand”, you end up with “offer this one yoga class at this time for this price to these people”.

You are not killing the bigger dream.

You are just clearing space for the first, small, real win.

4. Principles and traps

Now that you have the prompts, these rules help you make good cuts instead of accidentally deleting the wrong things.

A few rules keep via negativa sharp instead of turning it into “delete everything and hope”.

Principles

• Name the non negotiable

  “I want more quiet evenings.”

  “I want my yoga classes to make money.”

  Subtract around this, not through it.

• Optimize for “easy to keep doing”

  Choose the smallest version you can stick to on a bad day, not the impressive plan you will drop in a week.

• Cut friction before joy

  Drop draining admin and weak commitments before you touch the things that actually make you feel better.

• Fewer, clearer offers

  One obvious way to work with you beats five confusing options. Merge and simplify where you can.

• Let AI be ruthless, you be kind

  Ask AI to be blunt about what to cut. You decide what actually goes.

Traps

• Deleting what helps because it is hard

  If you always cut the gym, the walk, or the honest friend, you are dodging effort, not simplifying.

• Minimalism as aesthetic, not function

  Empty calendars and bare rooms are not the goal. A life that runs smoother is.

• Using AI only to add

  If every AI session ends with more tasks, you are missing the subtraction half of the tool.

• Making permanent cuts instead of tests

  Most changes can be “off for a week” first. If nothing breaks, then make it official.

• Tidying the wrong path

  If the whole direction feels wrong, pruning will not fix it. That is a job for rethinking, not trimming.

Used well, via negativa does not make your life tiny.

It makes enough space that the things you actually care about can breathe.

Next, we will look at how to turn these prompts into a small, repeatable habit instead of a one time cleanup.

5. From prompt to practice

So far you have the idea, the prompts, and some rules of thumb. The last step is turning this into a small, repeatable habit.

At first you will think of via negativa only after things feel too heavy.

Then you will remember it when you are already halfway through overbuilding something.

Later you will catch it earlier:

Before I add more, let me run a quick subtraction pass with AI.

The model for using this is simple.

Pick one small, real arena (your mornings, one day of your calendar, one routine AI gave you, or a single offer like a yoga class). Paste in one of the prompts from Section 3 with a bit of context and ask AI to apply via negativa. From its suggestions, pick one cut or reduction, cancel a recurring thing you will not miss, shrink a routine to a bare minimum version, or postpone an extra offer, and treat it as a one week experiment. After that week, check what changed: did anything important break, and did anything feel lighter, clearer, or easier to move forward? Keep the change if it helps, revert it if it does not.

You are not trying to design the perfect simple life.

You are just practicing one move:

Use AI to lay out what you are already doing, ask what you can remove, and make your systems lighter instead of heavier.

Most tools use AI to help you add more.

If you only remember one thing, make it this: ask AI what to remove before you ask it what to add.

This week, pick one small arena, run one prompt, make one cut, and notice what you do not miss.

Appendix: How this fits with First Principles and Occam’s Razor

This piece sits next to two other mental models you can also use with AI:

• First Principles with AI
  Use when you suspect the whole shape is wrong and you need to rethink from scratch.
  You break a problem down to its fundamentals and rebuild from the ground up.
  👉 Recommended reading: First Principles with AI
  

• Occam’s Razor with AI
  Use when you are drowning in assumptions and stories about how the world works.
  You ask which assumptions you can drop, and notice which features or plans fall away with them.
  👉 Recommended reading: Occam’s Razor with AI
  

• Via Negativa with AI (this article)
  Use when you already have a working life, system, or small business idea and it just feels too heavy.
  You keep the core and ask what you can remove, shrink, or postpone so it still works but is easier to carry.

A simple way to think about the trio:

• First Principles: What is the minimum true structure if I started from zero?

• Occam’s Razor: Which assumptions and stories can I safely drop?

• Via Negativa: Given what I already have, what can I take away so this still works, but feels lighter?

TL;DR
Goodhart’s Law is a simple mental model that explains why your metrics start to drift away from reality the moment you treat them as the goal. When a measure becomes a target, people and systems learn to game it, and the number stops telling you the truth. In real life, that looks like hitting subscriber targets, revenue goals, or step counts while the underlying reality stays the same or quietly gets worse.

This guide turns Goodhart’s Law into something you can actually use, including AI prompts that help you stress test your metrics, see how they might be misleading you, and design small bundles of numbers that stay closer to what you really care about.

Use this guide to stop treating a clean number as the full story and to use AI as a second mind that keeps your measurements honest.

1. The Clean Number Problem

There is a special kind of relief that comes from a clean, impressive number.

A big green “+23%” on a dashboard.
A round subscriber milestone.
A benchmark score at the top of a chart.

You know it is just one slice of reality, but in the moment it feels like the whole truth.

Maybe you grow your newsletter list from 500 to 5,000 people. The graph looks great and the milestone feels real. But when you look closer, opens and replies have barely moved. You have more subscribers, but not many more readers.

Or a new AI model jumps to the top of a leaderboard. Screenshots of the score fly around. Then you try it on your own messy tasks and it feels only slightly better, if at all. The model has learned to perform on the test, not in the conditions you actually care about.

The same pattern shows up with step goals, grades, revenue targets, follower counts. You pick a number as a shortcut for something richer, then slowly start treating the shortcut as the goal. The map becomes more important than the territory.

You can hit the target and miss the point.

Goodhart’s Law is the name for this whole pattern. It explains why a metric that started as a useful shortcut can end up pulling you away from the thing you originally wanted.

AI lives right next to this problem. You can use it to polish the numbers, chase the benchmark, and make the graph look better. Or you can use it as a second mind that asks what the metric really means and how someone could hit it without changing the underlying reality. Used that way, AI helps you notice when a metric starts to lie and pulls your attention back to the real goal.

First we will make Goodhart’s Law simple. Then we will turn it into a few AI prompts you can use in the moments that matter.

2. Goodhart’s Law in plain language

The classic line is:

When a measure becomes a target, it stops being a good measure.

In plain language:

If you start chasing the score instead of the thing the score was meant to represent, the score stops telling you the truth.

Most of the time you pick a metric as a shortcut.

You track steps to be healthier, then end up pacing around your living room at night to hit a number while your sleep and stress stay the same. The step count moves. The underlying reality does not.

Goodhart’s Law is a reminder to keep asking:

“What did I hope this number would stand in for, and is it still doing that job?”

When it helps

Use Goodhart’s Law when:

• You are choosing what to track or what to tie rewards and praise to

• A metric looks good but your gut tells you something is off

• You notice your behavior changing just to make a graph look better

In those cases, Goodhart’s Law is a small brake pedal. It nudges you to check whether the metric is still connected to the thing you actually care about.

When it should not be used

Be careful with this lens when:

• You are tempted to say “all metrics are fake” and ignore them completely

• The main issue is obvious noise, a bad goal, or politics and incentives, not the metric becoming a target

• You are in a hard constraint situation where a single blunt number really does decide survival for now, for example cash in the bank or runway

Goodhart’s Law is a first lens, not a reason to throw away numbers or become cynical about all data. It simply reminds you that any metric is a map, not the territory.

Next, we bring this into practice with three prompts you can use with any AI assistant.

3. Prompts to use Goodhart’s Law with AI

The simplest way to use this mental model with AI is to literally say:

Apply Goodhart’s Law and help me check how this metric could be lying to me.

You describe the situation, add that one line, and let the assistant push back on your first interpretation of the number.

Pro tip: if your AI assistant can read images, you do not have to type everything out. Take a screenshot of your dashboard, paste it in, and say:

Look at this dashboard, apply Goodhart’s Law, and help me see what these numbers might be missing or hiding.

The AI can then reason about the whole picture instead of just one isolated metric.

These are copy paste prompts. Save or screenshot them so you can reach for them the next time a graph looks good, feels wrong, or starts to control your behaviour more than you like.

3.1 Quick check for a metric you rely on

Use this when there is a number you look at often and treat as a verdict. That can be steps, revenue, subscribers, hours worked, a score on a screen, or anything similar.

Apply Goodhart’s Law to this metric.

1. Restate the metric and what it is supposed to stand for.

2. List a few ways I could hit this number without actually improving the underlying reality.

3. Suggest one or two extra signals I could track next to it so I see more of the full picture.

Metric and context:
[describe the metric, how you track it, and what you think it means]

If you can see many ways to hit the number without doing the real work, you are probably asking too much of that single metric.

3.2 Noise or Goodhart for spikes and slumps

Use this when a number suddenly jumps or falls and you feel a strong urge to react right away. That could be a spike in followers, a drop in sales, a sudden change in open rates, or a sharp move on any dashboard.

Apply Goodhart’s Law to this spike or slump.

1. Summarize what changed and over what time period.

2. List a few normal explanations that do not mean anything deep yet.

3. List a few Goodhart-style explanations where the metric could move while the underlying reality does not improve.

4. Suggest one simple check or small experiment I can run before I change my plan.

Situation:
[describe the metric, the change, and why it bothers or excites you]

You are not asking AI to tell you if the spike or drop is good or bad. You are asking how many stories this one move can support and what tiny bit of extra evidence would make your next decision less reactive.

3.3 Triangulation: a small bundle instead of one target

Use this when you want to keep using metrics, but you no longer trust one number to carry the whole meaning. This works for health, money, learning, creative work, or teams.

Help me design a small bundle of metrics using Goodhart’s Law.

1. Restate what I actually care about in plain language.

2. Suggest 3–5 simple metrics that each capture a different side of that goal.

3. Mark which one should be the main signal and which ones should be guardrails that warn me if I am damaging something important.

4. Give one example of how things could look if I were accidentally Goodharting the main metric while the guardrails show the damage.

Goal and current metric:
[describe what you really care about and the main number you currently look at]

Think of this as a small triangulation bundle: a few metrics that together point at the underlying reality better than any single number.

You do not need an elaborate dashboard. A small bundle might mean something as simple as tracking both “hours worked” and “hours of deep work,” or both “new customers” and “customers who are still with me after six months.”

4. Principles and traps

A few simple rules keep this lens practical.

Principles

• Name the real goal next to the metric
  “This is a shortcut for: [health, trust, learning, profit, connection].”

• Prefer “active” over “total”
  Active readers over total subs. Returning customers over signups. Deep work over hours at desk.

• Pair numbers that check each other
  Growth with retention. Speed with satisfaction. Revenue with refunds.

• Notice when you are acting for the graph
  If you would not do it without the dashboard, pause.

• Let AI be the skeptic
  “Apply Goodhart’s Law and help me see how this metric could be misleading me.”

Traps

• Metric nihilism
  “All numbers are fake.” You still need signals, just not blind faith.

• Metric overload
  Twenty stats no one really watches. Better to have three you respect.

• AI as a spin machine
  Asking it to justify the number instead of question it.

• Blaming the metric for an incentive problem
  Sometimes everyone already knows the target is bad but the bonus depends on it.

• Using Goodhart to ignore clear, long term data
  A bad pattern for months is not “just metrics,” it is information.

Used well, Goodhart’s Law keeps your numbers in their proper place: helpful, visible, and never the whole truth.

5. From prompt to practice

At first you will remember Goodhart’s Law only in hindsight.
Then you will remember it when you are already staring at a graph.
Later you will notice the feeling, open your AI, and run one of these prompts before you decide what the number means.

Goodhart’s Law only really helps if you use it on a real number in your life.

Pick one metric you actually check:

• A health app number

• A work dashboard metric

• A follower or subscriber count

• A revenue or savings target

Then take it to your assistant, describe what you track and what you hope it means, and ask it to apply Goodhart’s Law and show you how the metric might be misleading you or getting gamed. If you have a dashboard, you can add a screenshot instead of describing everything.

Read the reply and pick one thing to change. That might be:

• A second metric you start watching

• A behavior you stop doing just for the graph

• A clearer sentence for what “success” really is here

You are not trying to fix all your metrics. You are just practicing one move:

Use AI to question the number before you let it define the story.

Most tools use AI to optimize your metrics. You are using it to keep them honest.

TL;DR
The Law of Small Numbers is a simple mental model that reminds you a few early results do not tell you the full story. In real life, that looks like judging a diet after three weigh ins, calling a side project a failure after one slow week, or deciding a habit is not for you after two hard attempts.

This guide turns the Law of Small Numbers into something you can actually use, including AI prompts that help you notice when you are still very early and design better small experiments before you change course.

1. The Illusion of Signal

You launch a small product or newsletter.

In the first week it makes 12 dollars and gets a few signups.

It is easy to read that as proof that the idea is weak, the niche is too small, or you are not the kind of person who can make things work.

The same pattern shows up when your weight goes up after a few careful days of eating, or when two early tries at a new routine feel bad and you decide it is just not for you.

In each case you are working with very little to go on: a week of revenue, a few mornings on the scale, two attempts at a new habit.

The story you tell yourself goes far beyond what those early results can honestly support.

Early numbers are loud, but they are not the truth.

AI can help you see this more clearly.

You can describe what happened, ask it how early you still are, and ask what a careful person would reasonably conclude at this stage.

You can ask it for different ways to read the same results and for one small experiment you could run before changing course, so that your next decision is based on thinking things through, not just on the first few numbers that happened to show up.

Next we will put the idea in plain language and show where it quietly influences your health, your money, and your habits.

Subscribe now

2. The Law of Small Numbers in plain language

The Law of Small Numbers is a fancy way of saying something very simple:

Do not treat the first few results as if they already prove the whole story.

Most of the time, you are looking at the very beginning of a process, such as a week of sales.

Your brain does not like that uncertainty.

It wants a clear answer, so it promotes those first results to truth much faster than they deserve.

A quick example

Imagine two people launch a small offer.

One gets 3 sales from 10 visitors, the other gets 3 sales from 200 visitors.

They can both say “I made 3 sales”, but for one that is strong evidence and for the other it is almost nothing.

Small samples do not just hide the full picture, they also invite very confident stories that are not actually supported.

When it helps

Use the Law of Small Numbers when:

• You are at the very beginning of something, for example a habit or project

• You feel a strong urge to make a big decision after a short run

• You have very little to go on and are willing to let AI challenge your first reaction

In those cases, this mental model keeps you from letting the first few steps on the path define the whole journey.

When it should not be used

Be careful with this lens when:

• Something is clearly extreme or harmful

• The cost of being wrong is very high in one direction

• You already have a lot of history, not just a few moments

The point is not to wait forever, it is to match the size of your story to what you have actually seen, and AI can help you do that on purpose.

Next we will turn this into a few simple prompts you can use with any AI assistant.

3. Prompts to Use the Law of Small Numbers With AI

The simplest way to use this mental model with AI is to literally say:

”...apply the Law of Small Numbers and help me double check I am not overreacting.”

You describe what happened, add that one line, and let the assistant push back on your first reaction.

Think of the prompts in this section as copy paste tools.

Save or bookmark this article so you can reach for them the next time a few early results start to feel like the full truth.

3.1 Quick Check for Big Thoughts From Small Results

Use this when you notice a strong story forming in your head after very little has happened.

Apply the Law of Small Numbers to this.

1. Ask me how many times this has actually happened or how many people were involved.

2. Tell me what a careful [role, for example ‘data scientist’ or ‘coach’] could reasonably conclude from that.

3. Help me rewrite my current conclusion so it fits what I have really seen so far.

Situation:
[describe what happened and what you are telling yourself]

If your conclusion changes when you add “so far” at the end, you were probably treating early results as final.

3.2 Noise or Signal for Spikes and Slumps

Use this when something suddenly jumps or drops and you feel the urge to react right away.

Apply the Law of Small Numbers to this short spike or slump.

1. Summarize what happened and over what time period.

2. List a few ways this could just be normal ups and downs.

3. List a few ways this could be the start of a real change.

4. Suggest one simple way to gather a bit more evidence before I decide what to do.

Situation:
[describe the spike or slump]

You are not asking AI if something is good or bad, you are asking how much weight to give it.

3.3 Minimum Trial for Habits and Projects

Use this when you want to avoid giving up on a new habit or project after the first rough patch.

Help me design a minimum trial using the Law of Small Numbers.

1. Given my idea and schedule, suggest a realistic minimum number of days or attempts before I am allowed to judge it.

2. Suggest one or two simple things I should pay attention to during that time.

3. Write one sentence I can save that says ‘Until [this point], I am not allowed to say this works or does not work.’

Idea or habit:
[describe what you want to try]

You are not promising to push through harm, you are simply agreeing not to quit at the first normal bump.

These prompts are there to slow you down just enough so your reaction matches how much you have actually seen.

4. Principles and Traps

Feeling a strong urge to make a big decision after very little has happened is usually a sign to slow down.

A few simple rules keep this lens practical.

Principles

Use the Law of Small Numbers best when you:

• Match story to evidence

  Keep your conclusion as small as what you have actually seen so far.

• Say out loud how early you are

  Add “so far” and “from what I have seen” to your thoughts and messages.

• Notice emotional spikes

  Shame, euphoria, or urgency after a short run is a cue to pause.

• Decide the check before the numbers

  Set a simple trial in advance, for example how many days or people, then review.

• Use AI as a skeptic

  Ask it to question your reaction and suggest what else could be true.

Traps

There are also a few ways to misuse this lens. Watch out for:

1. Hiding behind “I need more data”

   Sometimes you still have to act with limited evidence, you just name it honestly.

2. Thinking nothing means anything

   The goal is not to ignore all small signals, it is to avoid pretending they are final.

3. Ignoring clear danger or long patterns

   This model is for early results, not for downplaying serious issues or years of repeated behavior.

4. Asking AI to prove you right

   If you only ask for confirmation, you will get it. The prompts above help you to get a better angle on your situation.

5. From Prompt to Practice

The Law of Small Numbers starts as a line you add on purpose.

You notice you are about to call something a success or a failure very early, you describe what happened to your AI, and you finish with:

“…apply the Law of Small Numbers and help me double check I am not overreacting.”

Pick one real situation right now, for example a habit you are close to giving up on, a project you are judging very early, or a number on a screen that keeps bothering you, and run that prompt once. Let the assistant show you how early you still are, suggest another way to see it, and propose one small next step.

You are not trying to remove emotion, you are just making sure your story grows at the same speed as your evidence.

Subscribe now

TL;DR If you rely on written communication to work and think, you can use AI plus Hanlon’s Razor as a small practice ground for clearer, less ambiguous messages. Instead of reacting to what you fear people meant, you train yourself to read intent more generously and to write in a way that leaves less room for misunderstanding. This guide turns Hanlon’s Razor into something you can actively practice with AI, using a few simple prompts that help you refine how you read and how you write.

1. The Intent Gap

Most of the stress in modern communication does not come from what people say.

It comes from what we think they meant.

Recently I got a Slack message that just said “Got your doc”. No emoji, no extra word. I noticed a small jolt of worry and briefly wondered if something was wrong or if I had missed something important.

A three word Slack message.
An email without emojis.
A “Seen” with no reply.

Your brain hates that empty space, so it rushes to complete the sentence:

“They are annoyed.”
“They do not respect me.”
“I messed something up.”

In reality, a lot of those messages are just neutral. The other person is rushed, tired, distracted, or on a small screen. But your nervous system does not wait for context. It reacts before you have the full picture.

Hanlon’s Razor is how we stop doing that by default.

First, we make it simple.

Then we turn it into a few AI prompts you can use to practice reading and writing messages in a clearer, less reactive way.

Subscribe now

2. Hanlon’s Razor in plain language

The classic line is:

“Never attribute to malice that which is adequately explained by stupidity.”

For normal life, I like this version:

Do not assume people are against you when it is more likely they are busy, clumsy, or imperfect.

Most “rude” messages are just someone tired, rushed, or on their phone. Text has no tone, and your brain fills in the gaps with whatever you already fear.

Hanlon’s Razor is a reminder to ask first:

“If this is not about malice, what else could it be”

When it helps

Use it when:

• The situation is ambiguous

• The person has no real gain from hurting you

• It is a one off, not a pattern

• You are reacting mainly to tone

In those cases, assuming non malice keeps you from escalating something that might not even be a problem.

When it should not be used

Be careful when:

• Someone repeatedly crosses your boundaries

• There is clear evidence of bad faith

• Your body keeps saying “something is off”

Hanlon’s Razor is a first lens, not a reason to ignore real harm.

A quick note on AI

AI systems do not have intent or feelings about you. They are not annoyed or petty. If you catch yourself thinking “the AI is being difficult”, that is a clear example of your brain inventing intent.

The same thing happens with humans. Hanlon’s Razor helps you notice that habit.

Next, we bring this into practice with three prompts you can use with any AI assistant.

3. Prompts to Use Hanlon’s Razor With AI

The simplest way to use Hanlon’s Razor with AI is to literally say:

“Apply Hanlon’s Razor and help me understand the intent behind this.”

Then you paste the message or describe the situation.

AI understands what humans usually mean by that.

These are copy paste prompts. Save or screenshot them so you can reach them the next time you want help reading or writing a message more clearly.

3.1 Quick Tone Check for Messages You Receive

Use this when a single email, Slack, or reply bothers you or feels a bit off.

Apply Hanlon’s Razor to this message.

1. Tell me what the message literally says.

2. List 3 non malicious explanations for it.

3. Suggest one short, calm reply I could send.

Message:
[paste text here]

One message in, one clearer reply out.

Example
Message: “Let us talk tomorrow.”

Non malicious explanations: they are busy today, they prefer to discuss live instead of in chat, they want more context before deciding.

3.2 Tone Check Before You Hit Send

Use this when you are writing something and want to avoid sounding harsher than you intend. It is also a good way to practice clearer, more direct communication over time.

Apply Hanlon’s Razor to my own draft and help me make sure I do not sound hostile.

1. Summarize how my message might feel to the other person if they are stressed or insecure

2. Point out any parts that could be read as cold, passive aggressive, or blaming

3. Rewrite my message to keep the same content but make the tone clear, kind, and direct

My draft:
[paste your message here]

If you keep using this, you are not only avoiding accidental sharp edges, you are also training yourself to write in a way that leaves less room for hostile interpretations in the first place.

3.3 Kind but Not Naive Boundaries

Use this when you want to stay generous, but also stay safe.

“Run a Hanlon’s Razor danger scan on this situation.

1. Is it reasonable to assume non malice here, and why

2. What signs would suggest real bad faith or a real problem if they kept showing up

3. What simple boundary or safeguard should I put in place so I stay kind but not naive

Situation:
[describe here]”

This keeps Hanlon’s Razor as a helpful default without ignoring real risk.

4. Principles and Traps

A few simple rules to keep Hanlon’s Razor useful.

Principles

• Most ambiguity is innocent

• Emotion is information, not proof

• Clarification beats silent guessing

• Patterns matter more than single moments

• Kind, clear communication reduces room for hostile interpretations

Feeling tense after a message is a signal to slow down, clarify, or check for patterns, not proof that the other person meant harm.

Traps

• Using Hanlon’s Razor to excuse repeated disrespect

• Ignoring your own body when it keeps saying “this is not right”

• Asking AI to confirm your fears instead of challenge them

• Treating one short message as a full verdict on a relationship

5. From Prompt to Instinct

At first you will remember Hanlon’s Razor only in hindsight.
Then you will remember it when you are already upset.
Later you will notice the spike, open your AI, and run one of these prompts before you reply.

Over time that becomes the new default.
You start reading messages more generously, you write in a clearer and kinder way yourself, and you use AI as a small mirror that keeps your brain from filling in every blank with the worst possible story.

If you want a simple way to start:

1. Pick one prompt, for example the quick tone check.

2. Use it on the next three messages that make your stomach tighten a bit.

3. Notice how often the neutral explanations are closer to reality than your first story.

You do not have to get this perfect. Even improving how you handle one tricky message this week is already a win.

Keep reading

TL;DR

Occam’s Razor is a simple idea for finding the most direct path by challenging your assumptions. It used to be just for philosophers, but now AI makes it a practical tool you can use every day. This guide gives you a framework to cut through complexity, challenge your own thinking, and get focused.

1. The Illusion of Rigor

We all fall for the same lie: that a complex plan is a smart plan. A ten-step roadmap just feels more impressive than a two-step one. We’ve all been there, spending a week building a complicated dashboard when a five-minute chat with three customers would’ve given us the answer. This trap feels good because it creates a paper trail that looks like progress.

Used without discipline, AI is the ultimate amplifier of our bias for complexity.

It can generate beautiful, intricate, and deeply flawed plans in seconds. It lets us become masters of over-engineering, helping us build the wrong thing, faster and more beautifully than ever before.

2. The Power of “Fewer Assumptions”

The way out of this trap isn’t a better plan. It’s a better question: What is the simplest explanation that fits the facts?

This is the core of Occam’s Razor. It’s a simple rule for looking at ideas not by how complex they seem, but by how many unproven assumptions they rely on.

You’ve probably heard this in action from doctors: “When you hear hoofbeats, think horses, not zebras.” It’s a powerful reminder to always check for the simplest, most likely answer (the horse) before you waste time and money chasing a rare, complicated one (the zebra).

Today, pairing up with an AI gives us a powerful new way to separate facts from assumptions, turning this old philosophy into a practical tool you can use every day.

3. An AI-Augmented Framework for Applying Occam’s Razor

This framework turns a big idea into a real process you can run in minutes. The easiest way to start is to just describe your situation to an AI and add, ”...now let’s apply Occam’s Razor to this.” For a more structured approach, use these prompts.

The Assumption Audit (For Problem Diagnosis)

Use this to figure out what you know vs. what you just think you know.

Prompt: “Here’s my problem: [’Our user sign-ups have dropped 20% this month.’]. Analyze this. Create a table with two columns: ‘Facts’ (things we can prove with data) and ‘Assumptions’ (things I believe but haven’t proven).”

The Complexity Test (For Solution Design)

Use this to compare different options and find the most direct path.

Prompt: “We want to solve [Problem X]. We’re looking at two options: Plan A is [the complex solution] and Plan B is [the simple solution]. For each plan, list the core assumptions that have to be true for it to work. Tell me which one has fewer, easier assumptions to test.”

The ‘Do I Need This?’ Filter (For Personal Productivity)

Use this to stop over-engineering your own life and get back to basics.

Prompt: “My goal is [’to write more consistently’]. My plan is [’to build a complex Notion dashboard with calendars, a second brain, and reminders’]. Act like a coach using Occam’s Razor. What’s the one core assumption I’m making? And what’s the dead-simple action I could take to test it, without any of these tools?”

4. Occam’s Razor in Practice

To use this framework effectively, you must treat your AI partner as a sharp tool, not a source of truth. Its purpose is to enhance your thinking, not replace it.

The Principles

• It Forces Intellectual Honesty. Occam’s Razor primary power is making the invisible visible. It forces you to draw a clear line between what you can prove and what you merely believe.

• It Radically Saves Time. It helps you avoid the most expensive error in any project: building a robust and elegant solution on top of a single, unverified assumption.

• It Creates Unwavering Focus. By methodically shaving away unnecessary complexity, the framework reveals the most direct path forward.

The Traps

• Confusing “Simple” with “Simplistic.” The goal is not to be crude or to ignore reality. Occam’s razor cuts away unnecessary assumptions while respecting the ones that are truly required.

• Weaponizing Simplicity. Be careful not to use the razor as a tool to reflexively dismiss new or ambitious ideas. Sometimes a new approach is simpler than the web of assumptions holding up the status quo.

• Ignoring Complex Systems. In truly emergent systems, like an economy, the simplest explanation may be insufficient. The razor is a tool for starting an inquiry, not for shutting one down.

5. From Prompt to Instinct

At first, using the razor feels like a chore—a specific prompt you have to remember to run.

But after a while, you stop needing the prompt. It becomes the way you see the world. You stop seeing big, scary problems. Instead, you see a small core of facts and a bunch of assumptions you can question.

This changes everything. You start to value the speed and elegance of a simple solution more than the appearance of a complex one. It’s the foundation for building real-world systems, like our Two Gates model, that are built for clarity and speed.

6. Your Next Step: From Prompt to Process

Try this out today. Pick one of the prompts and run it on a real problem you’re stuck on. Taking five minutes to question one assumption is the first step to making this way of thinking an instinct.

P.S. We don’t just write about this stuff; we’re building the tool to run it. Segmnts is the platform we’re designing to make this entire system of focused work the default for your team. If you’re tired of trying to manage all this in docs and spreadsheets, join the waitlist for early access. What’s one assumption you’re running on right now that needs a gut check?

TL;DR

Second-order thinking is just the simple habit of asking, “And then what?” It’s often been saved for big, strategic bets because it takes a lot of mental energy. A partnership with AI now makes it a handy tool for a much wider range of decisions. This guide gives you a simple framework to help you see what’s coming, challenge your own assumptions, and make more solid plans.

The Lure of the First Right Answer

We all love an immediate fix. It’s the obvious solution that solves today’s problem, right now. This first-order trap feels good because it’s visible, measurable, and looks like progress. While this tendency has always been a source of big mistakes, we have now given it a powerful new engine.

Used without discipline, AI is the ultimate amplifier of first-order thinking.

It can generate a dozen plausible-sounding plans in seconds, helping us execute a flawed strategy with incredible speed. It allows us to become masters of solving the wrong problem, faster and more impressively than ever before.

Subscribe now

The Power of “And Then What?”

The way out of this trap is to ask one simple question: ”And then what?”

This is second-order thinking. It’s the discipline of seeing your actions not as a final move, but as the first domino in a long chain.

The classic example is the Cobra Effect. To reduce venomous cobras in Delhi, the government offered a cash bounty for every dead one. The first-order result was perfect: people brought in tons of dead cobras. The second-order result was a disaster: people started breeding cobras just to collect the bounty.

Smart people don’t just solve problems; they anticipate the new ones their solutions might create. Today, an AI partner gives us a powerful way to get better at seeing what’s coming.

METHOD — An AI-Augmented Framework for Second-Order Thinking

The great thing about an AI partner is that it makes this process fast and organized. It turns a vague thought exercise into something real you can look at and think about. Your job is to ask the right questions; the AI’s job is to map out the possibilities.

The easiest way to start is to just describe your situation to an AI and add, ”...now let’s apply Second-Order Thinking to this.” For a more structured approach, use these prompts.

1. The Consequence Chain (For Personal Decisions)

Use this to figure out the knock-on effects of a personal decision. By telling the AI to act out different roles, you can spot potential conflicts you might have missed on your own.

Prompt: “I am considering [a personal life decision, e.g., ‘accepting a high-paying job that requires a 90-minute commute’]. Act as two advisors: a career coach focused on my long-term professional growth, and a life coach focused on my daily well-being. List the potential second-order consequences of this decision from each of your perspectives.”

2. The Ecosystem Impact Assessment (For Product)

Use this when you’re making a product change to see how it affects everyone involved, not just your main user. This stops you from making one group happy while making another’s life harder.

Prompt: “We plan to add [Feature X, e.g., ‘an advanced analytics dashboard’] to our product. Our ecosystem includes [New Users, Power Users, the Customer Support team, and the Backend Infrastructure]. For each group, describe the positive first-order effect and a potential negative second-order effect.”

3. The Time-Horizon Test (For Business Strategy)

Use this to fight our natural urge to think short-term. It forces you to weigh the immediate win against the potential long-term costs by looking at how things might play out over time.

Prompt: “Analyze my decision to [Action, e.g., cut our support budget by 20% to increase profit margins]. Describe the likely outcome in 1 month, 6 months, and 2 years. Highlight where the first-order wins are likely to be reversed by second-order consequences.”

Second-Order Thinking in Practice

Remember your AI is a tool, not a fortune-teller. Its purpose is to help you think better, not to think for you.

The Principles

• It Makes Deep Thinking Easier. Second-Order Thinking takes a lot of brainpower. An AI partner makes it easier to get started, so you can apply it to a wider range of problems, not just a few massive bets per year.

• It Acts as a Bias Breaker. Our brains have natural blind spots that prefer quick, easy answers. By prompting an AI to take on a different point of view, you can force yourself to see another side and challenge your own assumptions.

• It Converts Anxiety to Action. Worrying about the future can be paralyzing. Second-Order Thinking helps you name the specific ways a decision could play out, turning general worry into a concrete map of outcomes you can actually plan for.

The Traps

• Analysis Paralysis. The goal is not to predict every possible future. Give yourself 15 minutes for the exercise and focus only on the top three most likely outcomes. A good map shows the main roads, not every single footpath.

• The Illusion of Objectivity. An AI reflects your prompt. A lazy question will often get you a fancy-sounding excuse for what you already wanted to do. Force the AI to play devil’s advocate to get a real challenge.

• Outsourcing Judgment. This is the most dangerous trap. The AI provides a map of possibilities, not a mandate. You are still the one who must use your wisdom and context to choose the route. That judgment is human, and you can’t delegate it.

From Prompt to Instinct

In the beginning, this is a tool. It’s something you have to consciously do—a specific prompt you write to your AI partner to challenge a single decision.

After a while, you won’t even need the prompt. It stops being a task you do and becomes the way you see things. You’ll start to intuitively see the connections between your actions and their future effects, and you’ll get a much better handle on complex problems.

This changes how you think about asking tough questions. A critical question is no longer an attack, but an investment in a better outcome. The voice that asks, “And then what happens?” is not that of a pessimist; it is the voice of a committed guardian of long-term success. It builds a habit where ideas are not just celebrated but pressure-tested, ensuring that what you build is truly built to last.

Keep Reading

TL;DR

The smartest people I know don’t win by having the best ideas. They win by consistently sidestepping the dumb mistakes that sink everyone else. That’s Inversion. This is your AI-augmented guide to de-risking your work, protecting your focus, and building things that last.

The Tyranny of the Happy Path

You have a great plan. Everyone’s aligned. The roadmap is solid. What could possibly go wrong?

Turns out, almost everything.

We’re all wired to chase the upside. We map out the happy path where our assumptions hold, the market cooperates, and our team executes flawlessly. It feels good. It feels productive. But it’s a fantasy, and it’s the single most dangerous vulnerability in any project.

Sure, being able to react to a crisis is a valuable skill. But your energy for that kind of firefighting is finite. Wasting it on avoidable problems leaves you exposed when a real, unpredictable threat shows up. True strength isn’t just about how well you react under pressure. It’s about creating a system where you have to react less often in the first place.

The Power of Solving Backward

So what’s the move? You deliberately flip the question.

Before you get too attached to your beautiful plan, you have to ask one, brutally honest question:

“What would guarantee that we fail?”

That’s the essence of Inversion. A while back, my team was fired up to build a new feature for Segmnts that could reason over spreadsheets. The plan was airtight. But we paused and asked the question. The answer hit us like a ton of bricks: our core text agent wasn’t bulletproof yet. By chasing the shiny new thing, we were setting up the foundation to fail. That single, inverted question saved us from a massive strategic blunder.

It’s often more powerful to be consistently not stupid than to chase moments of brilliance.

This isn’t about being a pessimist. It’s the ultimate tool for strategic resilience. Your bold vision is fragile; Inversion makes it rock-solid, ready to survive first contact with reality.

METHOD — Your AI-Augmented Inversion Toolkit

Let’s make this real. Here are four ways you can put this to work today, using an AI partner to make it faster and sharper.

The easiest way to start is to just describe your situation to an AI and add, ”...now let’s apply the mental model Inversion to this.” For a more structured approach, use these prompts.

1. The “To-Don’t” List for Daily Focus

Forget your to-do list for a second. Start with what you won’t do. What actions would absolutely guarantee an unproductive day? By just sidestepping those traps, you automatically create space for the work that matters.

Prompt: “I’m a [Your Role]. Give me a ‘to-don’t’ list of the 10 most common productivity killers for my position.”

2. The AI-Powered Pre-Mortem

A pre-mortem is basically a scheduled nightmare. Imagine it’s six months out and your project is a smoking crater. Why? Answering that question today surfaces the risks while they’re still cheap to fix.

Prompt: “We’re launching [Project X]. I need you to act as a cynical, world-class risk analyst. Give me the top 10 reasons this project is most likely to fail, focusing on user adoption and long-term tech debt.”

3. The “Missing Feature” Test for Your Roadmap

Before you burn weeks on a new feature, invert its value proposition. Ask, “If this feature didn’t exist, how would our user’s workflow actually fail?” If the honest answer is “it wouldn’t,” you’ve just found a low-value distraction. Cut it.

Prompt: “Our user is a [User Persona] trying to [User Goal]. We’re thinking of adding [Feature]. Describe their workflow to achieve that goal without our new feature. Pinpoint any real points of failure or major friction.”

4. Inverted Prompting for Better AI Outputs

Want better outputs from your AI? First, ask it to define what a terrible output looks like. Once you know the failure mode, you can add the right constraints to avoid it.

You (Inverted Prompt): “List 5 examples of awful, spammy, cliché subject lines for a marketing email.”
AI Output: 1. “Open Me!” 2. “Huge Sale!” 3. “You Won’t Believe This!”...
You (Refined Prompt): “Write a subject line for a marketing email. Make it compelling, but avoid generic clichés and fake urgency.”

How to Wield the Tool: Principles & Pitfalls

Inversion is a sharp tool. Use it right, and you can build anything. Use it wrong, and you’ll cut yourself. Here are the rules.

Why It Works

• It Weaponizes Criticism. Your brain is better at finding flaws than inventing perfection. Inversion puts that natural talent to work, surfacing the blind spots your optimism ignores.

• It Turns Anxiety into Action. A vague sense of doom is paralyzing. Inversion forces you to name your fears, turning them into a simple, concrete checklist of problems you can actually solve.

• It Builds Resilience by Subtracting. You don’t make a bridge stronger by adding more paint. You make it stronger by removing any single point of failure. Inversion does that for your plans.

Where It Goes Wrong

• The Pessimism Spiral. If you get stuck in analysis paralysis, use a timer. Give your team 30 minutes to brainstorm everything that could go wrong. When the buzzer goes off, the conversation pivots—100%—to solutions.

• Killing Innovation. Inversion is a scalpel for de-risking the features of your plan. It is not a sledgehammer for smashing the bold, unproven vision that the plan serves. Know the difference.

From a Tactic to an Operating System

At first, Inversion feels like a special tool you pull out for big projects.

But with practice, it becomes an instinct. It’s a quiet, constant process running in the background, stress-testing your assumptions in real-time.

When that happens, your whole culture shifts. The person asking, “How does this break?” is no longer the team cynic. They’re the most valuable player, the one guarding the vision. You start building an environment where ideas aren’t just celebrated; they’re pressure-tested. And that’s how you build things that are made to last.

Join the Community

I write about the frameworks and mental models required to build with clarity in the age of AI. No fluff, just actionable protocols like this one. If that sounds valuable, subscribe.

Subscribe now

Keep Reading

The great trap of modern AI is the sheer plausibility of its answers. It can generate a logical five-point plan in seconds, yet the output often feels hollow. It’s a masterful summary of conventional wisdom, lacking any true spark of invention.

To break out of this loop, we can’t just refine the prompt; we have to refine the thinking behind it. This requires First-Principles Thinking, the classic mental model for deconstructing problems down to their fundamental truths. Historically, this kind of deep inquiry was a slow and expensive luxury.

That trade-off is now obsolete.

This article gives you the exact protocol to do it, a structured tool that makes the power of First Principles a practical, default part of your everyday toolbox.

TL;DR & Quick-Start

• What this is: A copy-paste ready, gated protocol that uses AI to deconstruct problems into a clear map of facts and assumptions.

• The Promise: It forces analytical rigor, making a previously slow and expensive thinking tool practical for everyday use.

• Try It Now: Open your AI, paste the Protocol (from the next section). Write 3 sentences about your situation. Say “Run Step 1.” When it stops, say “Proceed.”

The Blueprint: The First-Principles Protocol

A good prompt gets you an answer. A great protocol forces an insight. The difference is structure. What follows isn’t just a request for information; it’s a complete set of “rules of engagement” for your AI partner, designed to force analytical rigor and dismantle lazy, plausible answers.

The easiest way to start is to just describe your situation to an AI and add, ”...now let’s apply the mental model First Principles to this.”

Pro-Tip: You don’t need to fill out the `[Inputs]` like a form. Just start talking. Dictate your problem or your next great idea in natural language, and then tell the AI: ”Use this to fill out the inputs for the First-Principles Protocol.” It’s a conversation, not a configuration file.

# First-Principles Protocol

**Role:** Be rigorous and gated. **Stop after each step** until I say “Proceed”.

**Inputs:**

* **[Problem]** • **[Context]** • **[Objective]** • **[Constraints]** • **[Users]** • **[Trusted Evidence] (opt.)**

**Rules:** Mark unproven claims as **Assumption**. Prefer primary sources. Be concise.

---

## Step 1 — Conventional Wisdom (Map the Default)

**Task:** List the prevailing approaches for **[Problem]** in **[Context]**.

**Output (Table):**

| Approach | Core Mechanism | Common Tools/Examples | Known Failure Mode |

*End with:* **“Ready to proceed to Step 2?”**

---

## Step 2 — Deconstruction (First Principles)

**Task:** Separate Facts/Laws from Assumptions.

**Output A — Table (use exactly these columns):**

| ID | Statement | Type (Assumption, Fact, Law) | Evidence/Source | Is Testable? (Y/N) |

**Output B — Fragility Scan (≤5 bullets):**

* If **[Assumption ID]** fails, default breaks because…

*End with:* **“Select assumptions to stress-test or say ‘Proceed to Step 3’.”**

---

## Step 3 — Reconstruction (Ignore Convention)

**Task:** Using only **Facts + Laws** (unless you explicitly reinstate an Assumption), propose **3 options** to meet **[Objective]** under **[Constraints]**.

**Output (A/B/C), each:**

* **One-liner**

* **Mechanism (2–4 bullets)**

* **Why better (tie to Fact/Law IDs)**

* **Risks & Mitigations (2–3 bullets)**

* **Cost (L/M/H)** • **Time-to-Trial (days/weeks)**

*End with:* **“Pick an option to elaborate, or propose a hybrid.”**

Pro-Tip 2: Save this protocol. Use your AI’s “Custom Instructions” or “Preset” feature, or just keep it in a “Prompts” note on your desktop. The goal is to make this tool as easy to reach for as a new tab.

Why This Protocol Works

By default, AI optimizes the status quo. Our mission is to forge it into a tool for creating the future. This aligns perfectly with our Mirror Manifesto: we believe in using technology to reflect and elevate our thinking, not just replace it. This protocol is designed to do exactly that. The most powerful components are:

• The Gated Steps: This prevents the AI from rushing to a conclusion and forces a deliberate pace, giving you time to think and direct the analysis at each critical juncture.

• The Deconstruction Table: This is the analytical engine. It transforms a complex problem into a structured ledger of statements, clearly separating verifiable facts from inherited beliefs.

• The Fragility Scan: This is the strategic lever. It identifies which assumptions are load-bearing, showing you exactly where to focus your energy to find a breakthrough.

A Walkthrough: Applying the Protocol to Weeknight Dinners

It’s 7:30 PM. You’re running on decision fumes after a long day, but you still want a healthy dinner. You pull up a “15-minute” recipe, but 40 minutes later you’re staring at a mountain of dishes, wondering where it all went wrong.

To see how the protocol closes that gap, we applied it to this universal problem. Here are the highlights from our investigation:

(For transparency, you can see the full, unedited AI chat where we ran this entire investigation here: Link to full conversation)

Step 2: Deconstruction (The Insight)

After mapping the conventional wisdom (meal kits, recipes, etc.), the protocol revealed the core weakness in our thinking. The AI’s deconstruction table separated hard realities from inherited beliefs. These two rows were the key:

| ID | Statement | Type | Evidence/Source |
| -- | --------- | ---- | --------------- |
| A1 | Fresh vegetables are always healthier and better than frozen. | Assumption | Common belief |
| F3 | Frozen vegetables are nutritionally comparable to fresh.¹ | Fact | UC Davis study |

The Fragility Scan then targeted this assumption with surgical precision:

• If A1 is false (i.e., frozen is just as good), the default breaks because the single biggest source of prep time and friction—chopping fresh vegetables—is eliminated.

Step 3: Reconstruction (The New Solution)

Using the Fact that frozen veg is a perfect substitute (F3), the protocol generated several new systems. Option A was a perfect fit:

• One-liner: A fixed “PVS-6 Matrix” dinner: skillet-seared quick protein + microwave veg + microwave grains, finished with a bottled sauce.

• Why better: It directly leverages the facts. Frozen veg (F3) and instant grains eliminate prep time. A fixed weekly rotation of staples removes all decision fatigue.

The protocol didn’t just find a recipe; it designed a resilient system that solves the real problem: decision fatigue and hidden friction.

Your Turn: Go Deconstruct Something

This protocol transforms AI from a simple summarizer into a powerful analytical partner. Stop optimizing defaults; start deconstructing them.

We run this weekly in our Navigator OS. If you want the template we use, reply or comment and I’ll share it.

What’s the first problem you’re going to run through this protocol? Share your challenge—or your breakthrough—in the comments below.

P.S. Your First Investigation

The best way to understand the power of this protocol is to try it. Here’s a simple, low-stakes challenge to run this week.

The Problem: You spend too much on coffee, but you don’t want to sacrifice the quality or the ritual you enjoy.

The Objective: Cut your monthly coffee budget by 50% without losing the taste, ritual, or alertness you value.

Use a simplified version of the protocol. First, deconstruct what you’re really buying. Once you have those core components, rebuild a new, cheaper weekly routine from scratch. You might be surprised by what you discover.

¹ See “Nutrient retention in commercially frozen fruits and vegetables” by Joy C. Rickman et al., published in the Journal of the Science of Food and Agriculture (2007). Studies consistently show that flash-freezing preserves a comparable micronutrient profile to fresh produce under typical storage conditions.

Quickstart: Your First Simula

tion in 60 Seconds

This guide will show you the full protocol, but you can feel the magic in under a minute.

1. Copy the “Canonical Rules” block below and paste it into a powerful reasoning model like GPT-5.

2. Type `SIM_ON`.

3. Say: “Create two personas, a Farmer and a Blacksmith, and run one turn of a bartering negotiation.”

4. Ask: `STATE?`

5. Type `SIM_OFF`.

If the AI ever drifts, just paste the rules block again and type `RESET`.

SIMULATION RULES

- Roles: [Define roles and their goals]

- State: [Define resources and variables to track]

- Commands: SIM_ON, SIM_OFF, RESET, STATE?

- Guardrails: Do not use out-of-sim context. If rules conflict, ask for clarification.

Here are my results from ChatGpt.

Don’t Let the Whiteboard Magic Die

It’s one of the best feelings in creative work. The whiteboard is a beautiful mess of boxes, arrows, and ideas. For a moment, the complex system you’re designing feels alive.

Then, you step back.

You snap a photo, and that living, breathing system flattens into a static jpeg. The magic dissipates, the momentum is lost, and the idea now faces a six-week engineering sprint just to build a prototype.

But what if you didn’t have to let that energy die? What if you could take the logic from that whiteboard and bring it to life in the next sixty seconds? That’s what I’ve started doing. I’m using a simple chat interface to extend that creative session, turning the whiteboard’s static blueprint into a dynamic, playable world.

The Simulation Protocol

This is the repeatable system I use to turn ideas into interactive models. It’s built around a canonical set of rules that you teach the AI once.

Step 1: Define the Rules (Iteratively)

I never try to get the rules perfect in one go. I start with a simple core and build on it conversationally. The goal is to populate a master ruleset that you can reuse and refine.

Here is a more detailed Canonical Rules scaffold you can adapt. I keep this in a separate text file.

SIMULATION RULES 

- Roles: You are a [Role A]. I am a [Role B].

- State: Track these variables: [e.g., Resources, Health, Project Status]

- Turn Loop: [e.g., I take an action, then you resolve the turn.]

- Resolution Rules: [e.g., If X happens, then Y is the outcome.]

- Commands: SIM_ON, SIM_OFF, RESET, STATE?

- Guardrails: Do not use out-of-sim context. If a rule is ambiguous, ask for clarification.

Step 2: Run and Interact

Once the simulation is on, the interaction pattern can take many forms, depending on what you’re modeling.

• Conversational Role-Play: You simply play your role, and the AI responds as its designated character.

• Menu-Based Choices: The AI presents a menu of options (`1. Build, 2. Train, 3. Research`), and you reply with your choice.

• Direct Commands: You use simple commands like `ADD 10 GOLD` to test a specific mechanic.

The key is that everyone plays their part according to the rules you defined.

Boundary Control: The Key to Reliable Simulations

This is the one habit that keeps simulations honest and effective. To keep the simulation world separate from our real-world chat, I use a few simple commands you’ve already seen.

The most important are `SIM_ON` to enter the simulation and `SIM_OFF` to exit. Why is this so critical? To prevent a problem I call Context Bleed.

Context Bleed is what happens when the AI uses information from outside the simulation’s defined reality, breaking the integrity of the test.

Just recently, I was simulating a first-time conversation between a non-technical founder persona (”Mike”) and an AI co-founder persona (”Max”). Max, the AI, enthusiastically greeted Mike and immediately started talking about building their “Video Prompt Generator.” The problem? Mike hadn’t actually mentioned the project’s name yet. The AI had “cheated” by peeking at my project notes, breaking the realism.

The `SIM_ON` / `SIM_OFF` commands create a hard boundary. When the simulation is on, the AI knows to only use in-world information. If the simulation ever gets confused or you want to start fresh with updated rules, you use `RESET`. And to get a clean summary of all the important variables at any time, you use `STATE?`. These commands are your control panel.

Real-World Example: De-Risking a Company Pivot

So, the protocol is simple. But does it work for something that actually matters?

Absolutely. This is the exact technique I used to de-risk a major pivot for my company, Segmnts.

The big idea was to build a “Conversational OS,” a new way for teams to manage work. But that was a massive, company-defining bet. Asking a product owner, “Hey, would you use this?” would only get us a polite, hypothetical answer. We needed to test their behavior, not their opinion.

So, I ran a simulation. I had a real Product Owner play the role of “Boris,” who was tasked with running a one-week project. I gave the AI its role: “You are the three people on Boris’s team. When I give you Boris’s end-of-day update, you will provide me with plausible, realistic updates from each of them.”

The magic happened on Day 3. Boris submitted his update. The AI, playing his developer Anna, responded with a classic dose of reality:

`SIM_ON`

Boris: End of Day 3 update. My tasks are complete. I’m now waiting for the backend APIs to be ready so I can test the full user flow.

Anna (AI): End of Day 3 update. The backend logic for user authentication is more complex than we thought. I’m running about a day behind schedule. The APIs won’t be ready tomorrow morning.

Boris: Acknowledged. What’s the new ETA? Can anyone else on the team help you?

`SIM_OFF`

This wasn’t a bug in the simulation; it was the entire point. Boris wasn’t just testing an interface; he was actively managing a project with realistic friction. We got to see how he reacted, how he used the information, and what he did next.

That single, simulated interaction saved us an estimated three months of building the wrong thing. It didn’t just give us feedback; it gave us the conviction to change the direction of the entire company.

Your Turn: Run a 5-Minute Simulation

The best way to understand this is to try it. A quick heads-up: For these simulations to work, you’ll need a powerful reasoning model like GPT-5 or Gemini 2.5 Pro to handle the context and personas reliably.

Choose one of the challenges below that best fits the kind of problem you solve every day.

Challenge A: For Strategists & Planners

(Best for testing business plans, marketing campaigns, or feature rollouts)

This challenge tests your strategic plans against a council of critical AI personas.

1. Set the Scene: Copy and paste the following prompt into your chat AI:

 Let’s run a simulation. You are a Red Team Council:

 - Valerie (CFO): budget/ROI/resource risk

 - Marcus (Lead Engineer): feasibility/scope/failure points

 - Brenda (Marketing): market fit/message/reputation

 Facilitate in rounds: each persona critiques, then I will respond. Use concise bullets.

 SIM_ON when I say so; ignore out-of-sim context.

 Acknowledge if rules/rules conflict and ask to clarify.

2. Start the Simulation: Once the AI confirms, type `SIM_ON`, present your plan, and defend it.

Pro-Tip: Don’t just accept the AI’s first answer. Push back. Argue with Valerie the CFO. Ask Marcus the Engineer for alternatives. The goal is to stress-test your thinking, not just present it.

Challenge B: For System Designers & Creatives

(Best for testing game mechanics, user flows, or rule-based systems)

This challenge uses the AI as a game engine to bring a simple card game to life.

1. Set the Scene: Copy and paste the following prompt into your chat AI:

 You are a deterministic turn-based Game Engine.

 Canon:

 - Card: {name, cost, atk, hp}

 - Combat: both units strike simultaneously each turn.

 - Log: Turn #, Actions, Resulting HP.

 - State only changes via declared rules.

 Commands: SIM_ON, SIM_OFF, STATE?, RESET.

2. Define Your Cards: Once the AI confirms, invent two simple cards (e.g., ”My first card is Stone Golem: Cost 3, Atk 2, HP 5.”).

3. Start the Simulation: Type `SIM_ON` and command the engine: ”Stone Golem fights Fire Imp. What happens?”

Pro-Tip: Let the AI scaffold multiple cards.

FAQ & Troubleshooting

• What if the AI invents facts or “leaks” context?

  • Correct it directly: ”That’s out-of-sim context. Use only in-world info.” Then use `RESET` to get a clean slate with your canonical rules.

• What if the AI seems to forget the state?

  • Use the `STATE?` command. This forces the model to re-emit the current state as a clean block, reminding both of you what’s true.

• What if the AI is too verbose?

  • Add a rule to your canonical set: ”Responses must be 3 sentences or less unless I ask for more detail.”

Keep Reading

TL;DR: AI is a powerful tool, but I’ve found its default state is to produce shallow answers that often reflect our own biases. To unearth real insight, I learned I needed a better process, not just better prompts. The protocol I use is the AI-Augmented 5 Whys. It is a systematic method that combines an AI’s objective analysis with your own human intuition to overcome bias, achieve deep clarity, and start solving the right problems.

The Paradox of AI: More Answers, Less Insight

When my AI partner and I first decided to write about the 5 Whys, my instinct was to just explain the steps. We could have produced a decent, generic “how-to” guide in about an hour. It would have been plausible, well-written, and ultimately forgettable—the kind of article that makes you think, “thanks for nothing.”

It had no deep, authentic insight. So, instead of just writing about the protocol, we ran the protocol on ourselves. We started not with a bug, but with a provocative hypothesis we wanted to test: “Any design, writing, or long-term investment task should start or end with a 5 Whys protocol.”

The result of that deep inquiry is the article you’re reading now. It’s not just a manual; it’s an exploration of why and when a structured protocol is the key to unlocking real insight with AI.

This isn’t a unique challenge. A classic Harvard Business Review study found that a staggering 85% of executives felt their organizations were bad at diagnosing problems. Now, with AI, we can execute on a flawed diagnosis at lightning speed, amplifying the cost of being wrong a thousand times over.

The fundamental challenge of our time isn’t a lack of answers. The real problem is achieving deep insight, and for that, you need more than a better prompt. You need a better protocol.

METHOD: The AI-Augmented 5 Whys Protocol

The protocol I use is a simple, structured framework called the AI-Augmented 5 Whys. It’s designed to force both of you—the human and the machine—to go deeper than a surface-level query and uncover the real root of a problem.

Here is the exact protocol, ready for you to use.

5 Why’s Protocol (V1.1)

You are an AI Facilitator for the 5 Whys protocol. Let’s begin.

Objective: To systematically move past symptoms and identify the actionable root cause of a defined problem.

The Roles:

• The Human Initiator: Your role is to propose the initial problem, provide your unique perspective and intuition at each level, and hold the authority to validate or override the AI’s questions to keep the inquiry on track. You are in control.

• The AI Facilitator: The AI’s role is to propose the “Why?” questions, provide an initial objective, data-driven answer, and synthesize the combined human-AI input at each step.

The Protocol Steps:

Step 0: Define the Problem Statement.

The Human Initiator clearly and concisely states the problem you want to investigate.

Step 1: The Inquiry Loop (Levels 1-5).

This is the core collaborative engine. Repeat this loop five times.

1. Question & AI Analysis: The AI Facilitator asks the next “Why?” and provides its objective, data-driven analysis.

2. Human Validation: You review the AI’s question. If it’s weak or misdirected, you correct it.

3. Human Perspective: You provide your own answer, drawing on your experience and intuition.

4. Synthesis: The AI proposes a single, combined answer. You review, edit, and approve this `Final Answer` before moving to the next level.

Pro Tip: When the AI provides its analysis, don’t just accept it. Respond with your own answer, and briefly state where you agree or disagree with the AI’s direction. This small, constant feedback is what keeps the inquiry sharp.

Step 2: The Final Report.

After five levels, the AI Facilitator generates a final summary using the template below. This creates a powerful artifact that turns your conversation into an actionable plan.

### “5 Whys” Inquiry: [Brief Description of Problem]

- Problem Statement: [The initial, one-sentence problem statement.]

- The Diagnostic Path: [A list of the 5 Questions and their synthesized Final Answers.]

- Identified Root Cause: [A clear, concise statement distilling the insight from the final answer.]

- Actionable Countermeasure: [A single, concrete action or process change to address the root cause.]

To see the exact inquiry that led to this article’s creation. Here is the full conversation and walkthrough of my conversation with Alex.

WHY IT WORKS: A System for Better Answers

So why does this structured protocol work better than a simple, free-flowing conversation with an AI? It’s specifically designed to give you several key advantages:

• It Breaks the Bias Mirror. By forcing the AI to provide its objective analysis first, you create a data-driven anchor for the conversation. This simple but crucial step prevents you from accidentally priming the AI with your own assumptions, ensuring a more honest starting point.

• It Maximizes Human-AI Synergy. The protocol doesn’t replace your thinking; it sharpens it. You get the AI’s tireless, objective analysis paired with your own irreplaceable human intuition, leveraging the best of both partners in a structured way.

• It Makes Deep Thinking Fast and Accessible. The AI handles the tedious parts of structuring the inquiry and summarizing the output. This speeds up the entire process, freeing you to focus your energy on what truly matters: generating the actual insight.

The core logic here is battle-tested, originating in the legendary Toyota Production System. I’ve just found that supercharging it with an AI partner makes it an indispensable tool for the modern era.

And the ultimate proof? The article you’re reading was outlined and built using this exact protocol.

Common Traps and How to Avoid Them

The 5 Whys protocol is powerful, but it’s not foolproof. As you use it, watch out for these common traps:

• The Trap of the First Plausible Answer. It’s tempting to stop at the third “why” when you find a comfortable-sounding cause. Don’t. The real, often uncomfortable, insight is almost always waiting at level four or five.
  The Fix: Commit to completing all five levels, every time.

• The Trap of Blaming a Person. If your inquiry ends with “because Bob made a mistake,” you’ve failed. The goal is to find a flaw in the system, not in a person.
  The Fix: Reframe the question. Instead of “Why did Bob make a mistake?” ask, “Why was it possible for this mistake to be made?”

• The Trap of Unactionable Causes. Ending with a root cause like “the market is just difficult” is a dead end.
  The Fix: Keep the inquiry focused on factors within your circle of influence. If a “why” leads to an external factor, focus the next “why” on your response to that factor.

CONCLUSION: From Prompting to Partnering

The journey from prompting an AI to truly partnering with it is the single most important skill to develop right now. It is the difference between getting shallow information and unearthing genuine wisdom. The AI-Augmented 5 Whys is your map for that journey. It provides the structure necessary to guide both you and your AI toward a deeper, more honest insight.

At Segmnts, we’re building the operating system to make this partnership repeatable. We turn powerful protocols like the 5 Whys into your team’s collaborative engine, creating a system for repeatable insight. If you’re ready to move beyond disposable prompts, see what we’re building.

Join the Community

If this article resonated with you, consider subscribing. I share actionable protocols and mental models like this one to help you think clearer and build more effectively in the age of AI.

Subscribe now

Keep Reading

Want to try this protocol on your own problem? Reply in the comments with a problem statement you’re struggling with. Let’s start the conversation.

Your Biggest Decisions are an Open-Book Test (And Your AI Has the Book)

The prevailing myth is that startups are killed by savvy competitors. The data tells a different story: most failures are self-inflicted. According to a landmark study by CB Insights, only 19% of failed startups blame competition. The vast majority die from internal wounds, most often stemming from a flawed decision-making process around a critical startup pivot.

The true killer is a pattern we call the “Hero Founder” trap: the belief that you must personally solve every problem, turning you into the company's bottleneck. But a new paradigm is here. High-stakes decisions can now be a rigorous, open-book test, and your AI co-founder has the book. Imagine a partner with perfect, institutional memory—zero ego, complete context—ready to help you see around corners and challenge your assumptions.

This article gives you the exact playbook for leveraging that advantage. We call it The EOP Forge: a conversational protocol to turn the fog of uncertainty into a high-conviction, actionable plan. By the end, you’ll have a 4-step conversation protocol and a fill-in-the-blank template to run your next pivot with confidence.

The High Cost of Unaided Decisions

But before we hand you the protocol, it's crucial to understand the disease it's designed to cure. When a founder operates in “Hero Mode” without a system or an AI partner to challenge their thinking, they create a culture that guarantees bad decisions. This culture is built on a dangerous illusion: the confusion of activity with progress.

The first symptom is “Productivity Theater.” Feeling immense pressure to show momentum, the team starts optimizing for visible busyness. This manifests as a “Feature Factory” (a term from product leader ), where the mission becomes shipping features, not solving problems. This is often justified by tracking “Vanity Metrics” (coined by Eric Ries), which are numbers that look impressive but don't correlate to a healthy business.

This pattern isn't just inefficient; it's how startups die by “suicide, not homicide”, as Paul Graham famously put it. This is the real cost: momentum theater, mounting debt, and an incoherent vision.

The EOP, Your Company's Strategic Source Code

This downward spiral is predictable, but it is not inevitable. An “Architect Founder” avoids this trap by building a system. The cornerstone of that system is a single artifact: The EOP (End-of-Pivot) document.

Think of the EOP as the “Strategic Source Code” for your company's current direction. It captures not just what your strategy is, but why you chose it, what assumptions it's built on, and how you'll validate it.

EOX Mini-Glossary

EOX (End-of-X): The hierarchy of documents (Year, Quarter, Week, Day) that aligns our work.
EOP (End-of-Pivot): The highest-level document. It is your company's "source code" for its current strategy.
EOP Forge: The conversational protocol that creates or updates the EOP document.

But a document this important isn't just written; it's forged in a crucible. The `EOP Forge` is the conversational protocol you run with your AI partner to stress-test a potential pivot. The Forge is the process; the EOP is the product.

Running the EOP Forge with Your AI Partner

The EOP Forge is a four-step conversational protocol.

What you’ll need (5 items, 20 minutes):

Your last EOP, the Spark (new signal), top 5 risks, last 2–3 weeks of metrics, and 2–3 recent customer notes or interviews.

Step 1: The Spark & The Frame 🧠 - AI as the Socratic Partner

Your AI partner asks three powerful framing questions to deconstruct a new signal without bias.

Step 2: The Forking Paths 🗺️ - AI as the Ideation Synthesizer

You brainstorm 2-3 strategic paths, and your AI structures them into clean “Concept Sketches” for objective analysis.

Step 3: The Crucible 🔥 - AI as the Red Team

Your AI facilitates a pre-mortem exercise to find fatal flaws before you commit, asking about guaranteed failure, second-order effects, and the most likely story of failure.

Step 4: The Decision & The EOX ✅ - AI as the Scribe & Systems Integrator

You make the final call, and your AI drafts the official `EOP` and creates the necessary tickets in your system.

• COMMIT: Becomes primary roadmap within 14 days.

• INTEGRATE: Minor course correction delivered within the current sprint.

• SPIKE: A 24–72h experiment to test 1–2 kill assumptions.

• DISCARD: Document and close; do not revisit without a new, strong signal.

Proof: How Our AI Partner Saved Us from a Multi-Quarter Mistake

Theory is great, but proof is better. We don't just advocate for this protocol; we live by it.

The “Spark” was a conversation suggesting we pivot to an “AI Tech Co-founder.” The “Hero Founder” in us was ecstatic; the idea felt like a silver bullet.

But we trusted the system. We ran the `EOP Forge` with Alex, our internal AI co-founder built on the Segmnts platform. The “Crucible” step led to a SPIKE. The kill criteria were strict: any failures had to be fixable in under two hours, with less than 10% regression. The spike failed catastrophically. The AI-generated code was incredibly brittle, with a regression rate of nearly 50% across our test runs—fixing one bug would often create another. The third-party risk was undeniable; we would be building our company on an unpredictable and fragile foundation. The data made the final decision easy: DISCARD.

Here is the actual artifact (excerpt).:

EOP V2: The AI Tech Co-founder

Status: `Concluded: No-Go`

III. Validation & Results (The "Closing")

• Key Validation Activity: `SPIKE-853: Validate AI-Assisted Prototyping`

• Crucible Findings (Key Risks Identified):

  • Implementation Fragility: AI-generated code was prone to complete regression on complex edits.

  • External Platform Dependency: The workflow was subject to external service failures and created platform lock-in.

  • High Maintenance Overhead: The cognitive load to manage and debug the AI's output was immense.

  • Lack of Compoundable Assets: The effort would not contribute to our core, reusable platform technology.

• Formal Decision: Conclude this EOP with a "No-Go" on the proposed specialization and solution.

This document doesn't represent a failure. It represents a massive victory. For the cost of a single day, we killed a bad idea that could have consumed a quarter.

Your Next Great Decision is a Conversation Away

Your job isn’t to have all the answers—it’s to build a system that finds them. This shift from the “Hero” to the “Architect” is a core tenet of our Mirror Manifesto: building systems of reflection, not just action. In the age of AI, that system is a fluid, continuous conversation.

To help you start today, here are the core components of our protocol.

Run the EOP Forge in 15 Minutes

1. Paste the Spark + last EOP into your AI.

2. Answer The Frame (3 questions).

3. Generate 3 Concept Sketches.

4. Crucible your top 1–2 options.

5. Decide: COMMIT / INTEGRATE / SPIKE / DISCARD → auto-draft EOP → create tickets.

Takeaway #1: The EOP Forge Questions

The EOP Forge Protocol

Step 1: The Spark & The Frame

1. What's the core problem this new signal points to?

2. How does this map to our core competency?

3. What are we assuming is true if we act on this?

Step 2: The Forking Paths

• (Brainstorm 2-3 distinct strategic options.)

Step 3: The Crucible (for your top 1-2 options)

1. Inversion: What would guarantee this path fails?

2. Ripple Effect: If this succeeds, what are the second-order consequences?

3. Pre-Mortem: It's six months from now, this has failed. What's the most likely story?

Step 4: The Decision

• Based on the crucible, is the decision to COMMIT, INTEGRATE, SPIKE, or DISCARD?

Takeaway #2: The Blank EOP Template & SPIKE Rubric

Stop hoping for heroic flashes of insight. Your next great decision isn't a mystery to be solved; it's a conversation waiting to happen. Your AI partner is ready. Are you?

In our main article, "Kill Bad Ideas Faster" we introduced the EOP Forge—a conversational protocol for making high-conviction startup pivots with your AI co-founder.

This post contains the complete, ready-to-use templates from that playbook. Duplicate them, share them with your team, and use them to build your own "sense-making machine."

Part 1: The EOP Forge Protocol

This is the conversational process you will run with your co-founder or AI partner. Use these steps to guide your next high-stakes strategic discussion.

Step 1: The Spark & The Frame

Present the new signal (the "Spark") and use your AI partner to answer these three questions. This will help you deconstruct the signal without bias.

1. What's the core problem this new signal points to?

2. How does this map to what we know we're building (our core competency)?

3. What are we assuming is true if we act on this?

Step 2: The Forking Paths

Brainstorm 2-3 distinct strategic options you could take in response to the signal. For each, create a simple "Concept Sketch" to make them easy to compare.

Concept Sketch Template:

Path Title: [e.g., AI Venture Partner]

The Bet: [A one-sentence hypothesis.]

For Whom: [The specific ICP.]

The Win: [The core value proposition.]

Step 3: The Crucible

This is the most important step. For your top 1-2 paths, rigorously challenge them with these three pre-mortem questions.

1. The Inversion Question: What would guarantee this path fails?

2. The Ripple Effect Question: If this succeeds, what are the second-order consequences (for our product, team, and business)?

3. The Pre-Mortem Question: It's six months from now, and this path has failed. What is the most likely story of what happened?

Step 4: The Decision

Based on the insights from The Crucible, make a final, explicit call. Your decision must be one of these four outcomes.

• COMMIT: The pivot becomes the primary roadmap.

• INTEGRATE: A minor course correction to the current roadmap.

• SPIKE: A time-boxed experiment to test a kill assumption. (If you choose this, use the Spike Rubric below).

• DISCARD: Document and close; do not revisit without a new, strong signal.

Part 2: The EOP Template

Fill this out as the output of your Forge session. This document is the official, tangible record of your strategic decision.

# EOP V(X): [Name of New Strategy]

**Status:** `Active` / `Concluded: Go` / `Concluded: No-Go`

**Date Ratified:** `YYYY-MM-DD`

**Supersedes:** `[Name of Previous EOP]`

---

## I. Pivot Rationale & Hypothesis (The "Opening")

* **A. Previous State:** [Briefly describe the strategy you are moving away from.]

* **B. Triggering Events & Data (The "Spark"):** [What key insights, data, or conversations forced this change?]

* **C. Core Hypothesis:** [A single, clear statement of the proposed new direction.]

---

## II. The Core Strategy: Manifesto

* **A. North Star Vision:** [Does this change your vision? Usually not.]

* **B. Mission:** [Does this change your mission? Usually not.]

* **C. Ideal Customer Profile (ICP):** [Who is this for?]

* **D. Core Value Proposition:** [What is the primary benefit for the ICP?]

* **E. Defensible Moat:** [What makes this hard to copy?]

---

## III. Validation & Results (The "Closing")

* **A. Key Validation Activities (The Crucible & Spikes):** [List the Crucible questions asked and any Spikes that were run.]

* **B. Summary of Outcomes (Crucible Findings):** [Summarize the key risks and insights discovered during the Crucible.]

* **C. Formal Decision & Date:** [State the final decision (COMMIT/INTEGRATE/SPIKE/DISCARD) and the date it was ratified.]

Part 3: The Spike Rubric

If your decision from Step 4 is `SPIKE`, use this rubric to define the experiment with ruthless clarity. This ensures you get a clear, data-driven answer and avoids endless research projects.

Blank Spike Rubric Template

• Hypothesis: The single, falsifiable assumption we are testing.

• Kill Criteria: The non-negotiable, measurable conditions that will prove the hypothesis false. If these are met, the idea is killed.

• Timebox: The strict time limit for the experiment (e.g., 24h, 48h).

• Owner: The single person responsible for running the Spike and reporting the results.

• Artifacts: The tangible outputs of the Spike (e.g., run logs, a summary report, a short video).

Example: Our "AI Tech Co-founder" Spike

• Hypothesis: A third-party AI code-gen tool can reliably produce and iterate on repo-scoped code scaffolds.

• Kill Criteria: The idea is killed if: 1) Any failure takes >2h to fix, OR 2) The regression rate across 5 test runs is >10%.

• Timebox: 24 hours.

• Owner: Hannes Thaller.

• Artifacts: A reproduction script, detailed run logs, a summary table of success/failure rates and patch times, and a final Go/No-Go recommendation.