THE PROMPT VAULT
The most expensive thing I ever built was a beautiful plan.
Twenty two pages, a real financial model, three scenarios, the whole production. Took a month. Everybody who read it nodded. We executed it, and about ninety days in it became obvious that the entire thing rested on one assumption nobody had ever said out loud, and that assumption was wrong.
Not slightly wrong. Wrong in the way where the plan doesn't need adjusting, it needs to not exist.
Here's what stings. We'd have found it in about forty minutes if a single person in that room had been assigned to attack the thing instead of improve it. Nobody was. Everybody was in build mode, and build mode is structurally incapable of noticing that the foundation is sand, because build mode's entire job is to keep building.
That's the gap these prompts fill. AI is genuinely excellent at attacking a plan, better than most humans, for a reason that's almost funny: it has no ego investment in your idea, no career riding on the outcome, and no interest in staying on your good side. Ask a team member to poke holes and they'll poke gentle ones. Ask a model and it'll go for the throat if you tell it to, and it'll do it in ninety seconds.
Seven prompts. Run them in order on anything you're about to spend real money or a real quarter on. A new offer, a hire, a price change, a market, a partnership, a build.
Start here, always. This one catches the twenty two page disaster.
Below is a plan I'm about to execute. Read it and identify the assumptions it depends on that are never stated anywhere in the document. I'm not asking about the assumptions I've written down. I'm asking about the ones so obvious to me that I forgot they were assumptions.
For each, rate how confident I appear to be versus how confident the available evidence would justify. Flag any assumption where those two numbers are far apart, and tell me the cheapest way to test it before I commit.
Plan: [paste]
The word "cheapest" is load bearing. You want a test that costs a hundred dollars and two days, not a test that costs a pilot program. If the model comes back with an expensive test, push it: "give me a version of that test that costs under $200 and takes under a week."
Most of the time the cheap test exists and you just hadn't thought of it because you were already emotionally past the question.
Two: the six month postmortem
Classic pre-mortem, and it works because it changes the question from "will this fail" to "how did it fail," which your brain answers much more freely.
It's six months from now. I executed the plan below and it failed badly enough that I regret it.
Write the honest internal postmortem. Not a polite one. Include: the root cause, the three earliest warning signs and roughly when each would have appeared, the decision point where we could have changed course cheaply, and the thing everybody privately knew but nobody said in the original planning meeting.
Plan: [paste]
That last item is the one worth reading twice.
Take the early warning signs and turn them into actual tripwires with dates. If sign one shows up in week three, you want a calendar reminder in week three telling you to go look. A warning sign you never scheduled a check for is not a warning sign. It's a thing you'll notice in hindsight.
Three: the competitor's memo
This one's my favorite because it consistently produces the most uncomfortable output.
You run my strongest competitor. You just got a full briefing on the plan below because it leaked.
Write the internal memo you'd send your leadership team. Be honest, the way people actually are internally. Where should we be worried? Where is this obviously going to struggle? What's the cheapest way for us to blunt it? What are they missing about this market that we know?
Plan: [paste]
Something about the framing gets you a much sharper read than "what are the weaknesses of my plan." Weaknesses is a request for a balanced assessment. A competitor memo is a request for the truth as someone with opposing interests would see it, and opposing interests produce clarity.
The "cheapest way to blunt it" line is where the real value hides. If a competitor can neutralize your entire advantage with a two week feature or a price match, you didn't have an advantage. You had a head start.
Four: the second order consequences
Every plan gets evaluated on whether it works. Almost nobody evaluates what happens if it works.
Assume the plan below succeeds completely and hits every target. Now walk me through the second and third order consequences.
What breaks under the new volume? Which processes that currently work fine stop working? What does this commit me to that I haven't accounted for? Who on my team is now doing a job they didn't sign up for? What does my cost structure look like at 3x the current volume, specifically where it changes shape rather than just scaling?
Plan: [paste]
I've seen more businesses damaged by plans that worked than plans that failed. Failure is loud and you fix it. Success that outruns your operations is quiet, and by the time it's obvious you've got angry customers, a burned out team, and a delivery model that only functioned at the old size.
Answer this one before you launch, not after the good problem arrives.
THE AI WORKFLOW BLUEPRINT • $47
The complete vault, organized by decision type, plus the twelve core workflows that run the operational side of a small business. Every prompt tested on real plans with real money behind them, not written to look good in a newsletter.
Five: the resource honesty check
Plans fail on capacity far more often than on strategy, and capacity is the thing owners are most reliably wrong about.
Here's my plan and here's my actual situation: [team size, roles, hours genuinely available, current commitments, cash position].
Tell me what this plan requires in hours per week, broken out by person and by phase. Then compare that against what I actually have available, accounting for the fact that existing work does not stop while new work starts.
Where's the shortfall? If I can't hire, what would I have to stop doing to free up the capacity? Be specific about what gets dropped, not general about "prioritization."
The follow up question is the one people avoid, so ask it explicitly: "What would have to stop?" If there's no honest answer to that, the plan doesn't fit and you're planning to run it on evenings and weekends, which is a plan to run it badly and resent it.
This one lands harder when you have real numbers instead of your impression of your numbers. I run Rize in the background specifically so that when I say "I have ten hours a week for this," I'm reading a measurement instead of making a guess. The guess is always optimistic by about forty percent, in my case reliably so.
Six: the customer's actual reaction
You wrote the plan from inside your own head, which means you know why every piece of it makes sense. Your customer does not have that context and never will.
Play three of my customers reacting to this change, and make them realistic rather than agreeable.
Customer A has been with me three years and likes things how they are.
Customer B is new, price sensitive, and comparing me to two alternatives right now.
Customer C is my ideal customer, the profile I want more of.
For each: their honest first reaction, the objection they'd voice, the objection they'd have but not voice, and what would have to be true for them to be genuinely happy about this.
Change: [paste]
The unvoiced objection is the whole reason to run this prompt. Voiced objections you'd have handled anyway. Unvoiced objections are the ones that show up sixty days later as churn you can't explain.
If your team records customer calls, you can make this dramatically better by feeding in real language from real conversations. I pull specifics out of Fathom transcripts for this, because a synthetic customer built from your actual customers' words is a much better simulation than one built from a description you wrote.
Seven: the kill criteria
Last one, shortest one, most skipped one.
Based on the plan below, write the specific conditions under which I should kill this and walk away. Give me hard numbers and hard dates, not soft language.
Format: "If [measurable thing] has not happened by [date], stop." Give me three of these, covering the early signal, the midpoint, and the final call.
Then tell me what I'm likely to tell myself at each of those moments to justify continuing anyway.
That final line is the point of the entire prompt. You already know you're going to rationalize. Having the rationalization written down in advance, in your own words, before you're emotionally attached to a project you've spent four months on, is one of the few genuinely reliable defenses against sunk cost.
Print it. Tape it somewhere. When you hit the date and you're mid rationalization, you'll recognize your own excuse on the page and it gets a lot harder to use.
One thing these prompts cannot do
Worth being straight about the limits, because the failure mode here is treating the output as a verdict.
A model attacking your plan is working from what you fed it plus a general understanding of how businesses tend to break. It does not know your market's specific weirdness, your customers' actual loyalty, or that the competitor everybody fears has been quietly falling apart since March. You know those things. It doesn't.
So read the output as a list of questions worth answering, not a scorecard. When it flags something you know is wrong, say so and move on. When it flags something that makes you defensive, sit with that one a while longer, because defensiveness is usually a signal that it landed somewhere real.
The value isn't in the model being right. It's in seven structured attacks arriving in forty minutes instead of never.
Running the set
Forty minutes for all seven if you move. Do it in one sitting, before you tell anybody the plan, because once you've announced it publicly you're defending rather than examining.
You will not kill most plans this way. That's not the goal. The goal is that the plan you execute is the version that survived seven attacks, which is a materially different plan than the one you walked in with.
The twenty two page disaster I mentioned would have died at prompt one. Forty minutes instead of a quarter and a lot of money. That math has never stopped being worth it.
THE AI BUSINESS ACCELERATOR • $97
Eight weeks of building the systems, running the plans through this exact gauntlet before you spend anything, and installing the tripwires that tell you early when something's off. Bring the plan you've been circling. We'll break it, then build the version that holds.
Jordan
The AI Newsroom | Jordan Hale | ainewsroomdaily.com