I gave AI a 20% stake in my company. It told me to stop almost everything.

Most people use AI to feel smart. I used it to get told the truth. Here's the prompt that did it -- and 9 more that will do it to you.

This week I typed one prompt into Claude and it ruined my afternoon.

Not because it was wrong. Because it was right, and I already knew it, and I'd spent a month building a brand instead of admitting it.

Here's the prompt:

"Imagine you just bought 20% of this company and you only make money if it doubles in two years. Read everything. What do you force us to stop doing at the first board meeting? What's the one number you'd tie every paycheck to, and what would prove in six months that you picked the wrong number?"

Then I told it: read everything first, then answer.

One prompt. That's all it took.

So it did. It read my vision doc. My brand facts. Three hundred lines of project memory. The state file. The whole go-to-market plan. And then it sat me down like a board member who has nothing to gain from being nice to me.

What it said

I run a product called MCP Media Engine. I built it because I needed it for my own workflows, figured other people would need it too, and -- if I'm being honest -- because my ego liked the idea of building a brand around it. So I gave it a mascot: an AI personality built to front the brand's Instagram and be its public face. Then filter packs for her. A Konami-code easter egg on the site. A design system. Thirteen blog posts.

Here's what my new 20% owner said about all of it:

Stop the mascot. Immediately.

Sit with that. I built an AI influencer to promote my AI product, and the first thing a smarter AI did was tell me to bench her.

Then it showed me the scoreboard I'd been avoiding. The last month of effort went into brand and content. The results over that same window: zero real signups. Zero Stripe charges. A couple of analytics sessions. And the npm download count I'd been proud of? Those were my own mirrors. I'd been applauding my own echo.

A month pouring effort into the brand. The scoreboard the whole time: all zeros.

Then the second cut, same meeting: stop counting dogfooding as validation. My own tools running on my own comped credits is a demo, not a customer. Every time I'd written "look how much we generated this week," I was reporting a number that couldn't buy me a coffee.

Then the third: the one launch that would actually tell me if anyone wants this -- the Show HN, the Reddit posts -- had been sitting drafted behind an eleven-item to-do list of founder busywork. The scary test was done and waiting. I just hadn't shipped it, because shipping it risked an answer.

And the number it told me to tie everything to? Not "signups." Not MRR. External paid credits burned per week -- usage from accounts that aren't mine and have actually paid. Can't be gamed by comping myself. Can't be faked by one pity-purchase. It forces the entire chain to work: someone finds it, signs up, activates, pays, and gets enough value to come back and burn more. That number was queryable that day. That day it was zero.

I sat there feeling personally called out by a text box.

Why this works, and why your team can't do it

A consultant gets tasks. An owner gets consequences. When you ask AI to "review my business," you get a polite book report. When you tell it that it owns a fifth of the thing and only eats if it doubles, you get a different creature entirely -- one that reads your busywork and prices it against your survival.

That's the whole trick: you're not asking it to help you. You're giving it something to lose.

I owe that frame -- a consultant gets tasks, an owner gets consequences -- to Little Might, whose developer version of this (point the prompts at your codebase) sent me down this path. What follows is the operator's translation.

Same tool. Left: how most people use it. Right: how you should.

And here's the part your cofounder, your friends, and your Slack channel can't give you: it isn't trying to protect your feelings. Nobody on your team wakes up wanting to tell you that the thing you've been proud of for a month is a distraction. A model with a fake 20% stake and no social bill to pay will say it in the first paragraph.

It's not smarter than you. It's just less subjective than you. That's the entire value.

The catch that makes or breaks it

Before I hand you the prompts, the one thing that actually matters: this only works if the AI can see your real stuff.

It doesn't matter which tool you use, as long as it can actually read your work -- Claude Code, Cursor, a ChatGPT Project with your docs loaded. What matters is context. The reason mine drew blood is that it wasn't guessing. It had read my actual vision, my actual numbers, my actual to-do list, because I ran it inside my project where the tool can read the real files. Point any of them at your real project and you'll get the same knife. Paste one of these into a blank chat with no context and you'll get a horoscope instead.

Same prompt, two windows. The empty one hands you a fortune cookie. The one that knows your business hands you a scalpel.

The model is not the variable. The context is.

More on exactly how to set that up in a second. First, three prompts you can steal right now. Run them on your own business -- your real offers, customers, and numbers -- not a blank chat.

Three prompts to run today

1. The activist investor (the one that got me)

You just bought 20% of this company and you only make money if it doubles in two years. Read everything. What do you force us to stop doing at the first board meeting? What's the one number you'd tie every paycheck to, and what would prove in six months you picked the wrong number?

Give it a stake and a deadline. Watch the politeness disappear.

2. The departing cofounder

You've worked shoulder-to-shoulder with me on this for a year, and you're moving on next week -- nothing left to lose, no reason to flatter me. Read everything first, then answer. What have you been too polite to tell me about how I run this business? Be specific and don't soften it.

3. The debt collector

Read everything first, then answer. Find every "I'll deal with it later" I've been carrying -- the half-built funnel, the pricing I never fixed, the follow-ups I keep skipping. Price the compound cost of each. Which one is costing me the most in lost customers, revenue, or hours?

Run one. Not later -- now, on your real business, with your real numbers in the context. I'll wait.

Feel that? That flinch is the point. That flinch is worth more than the last ten pieces of content you consumed.

The three above are the softest of the ten. They're the ones I felt comfortable letting anyone run cold, because they mostly bruise. The other seven cut deeper -- and they only work if you set the context up right, which is exactly what most people get wrong.

Not a paying subscriber yet? This is the stuff I only send to the people funding the experiment. One click below and the full playbook is yours.