Senior year of high school, a paper came due that was worth half my grade. Around twenty-five pages. I had, of course, written none of it, because I never did homework in my life. So I did what half the world would learn to do a decade later: I handed the assignment to an AI.
It was 2015 and the tool was primitive, one of those early essay generators, and it produced twenty-five pages of grammatical, confident, total nonsense. Word salad in a suit. I turned it in anyway. A merciful English teacher passed me, barely, in what I can only describe as an act of grace, and without her I would not have graduated high school. If you are reading this: thank you. And I am sorry.
Here is why that story matters beyond the comedy. The question everyone asks now is how to use AI in a business, and I got the answer early, at full tuition: use AI as leverage on work you understand, never as a substitute for understanding. Delegate the labor. Never delegate the judgment. AI multiplies operators and exposes passengers, and it has been doing that since it could barely write a sentence.
What 2015 me got wrong
The machine did not fail me. Look closely at what actually happened: I outsourced the labor, which was fine, and the judgment along with it, which was fatal. I could not grade the output, because I had never learned the material, so I shipped twenty-five pages I could not evaluate to a customer, my teacher, who could. The tool’s weakness was not the problem. My inability to check its work was.
Now watch modern businesses run my senior year on a loop. AI-written proposals sent unread. AI answers pasted to customers containing confident errors. AI marketing copy shipped by people who cannot tell strong from slop. The tools are a thousand times better than my essay generator, which makes the failure mode more dangerous, not less, because the nonsense no longer looks like nonsense. In 2015 the gibberish announced itself. Today it wears a suit that fits.
The tools grew up; the lesson didn’t change
Today’s AI genuinely writes, analyzes, codes, and answers well. The bottleneck moved. Output quality is no longer the scarce thing; direction and evaluation are. Which means the leverage equation flipped into the operator’s favor: a person who deeply understands their business can now multiply themselves absurdly, while a person who understands nothing can now produce impressive-looking nothing at industrial scale.
Treat the machine exactly like a talented new hire, because every delegation rule I have ever written applies to it verbatim. Define what good looks like before you ask. Give it examples, context, and constraints, not vibes. Inspect the early output tightly, then loosen as trust accumulates. And remember the Postmaster rule: everything you fail to specify becomes the machine’s judgment call, executed at full speed, in your name. The tools will keep improving. The rules will not change, because the rules were never about the tools.
Where AI actually pays in a small business
Concretely, the highest-return uses I see for a small operation, in rough order:
Drafts and variations. First versions of emails, proposals, product descriptions, ads. The machine produces volume; you grade and finish. Ten graded drafts beat one perfect blank page every day of the week.
Documentation and SOPs. Describe a process out loud, get a checklist back, fix it once, and you have built the delegation infrastructure most small businesses never write. The machine is the cheapest process-documenter ever invented.
Research and summarization. Competitor teardowns, review mining, contract first-reads, learning a new domain fast. Hours become minutes, as long as you spot-check what matters.
Customer service first drafts. The machine drafts, a human grades and sends. Care stays human; typing gets cheap. Never let it answer unsupervised in your name.
Estimating and quoting. I was pricing lawns off satellite images at thirteen because the leverage was free and nobody else looked. The modern version of that gap is AI, sitting in the open, mostly unused by your competitors. Same opportunity, bigger.
What you can’t hand it
Now the boundary, learned the hard way in 2015 and confirmed every year since. You cannot hand the machine your judgment: what good means in your business is a standard only you can set. You cannot hand it your taste, your read on a customer, or the decision of which game to play in the first place; a compass cannot be delegated, and AI is the fastest clock ever built. And you can never hand it accountability. When the confident nonsense ships, the customer does not blame the tool. Your name is on the paper. It is always your name on the paper.
The rule that survives every model release: AI raises the ceiling on execution and raises the price of bad direction, simultaneously. Faster horses, same rider. The rider still decides where the race goes.
Watch the market split into two failure modes around that rule. The refuser bans the tools on principle and pays full price for typing while competitors pay wholesale; that is a margin decision disguised as a values decision, and margin decisions lose to arithmetic eventually. The passenger does the opposite: ships everything the machine produces, ungraded, at volume. That is my senior year, industrialized. Both camps lose to the same person: the operator standing between them, delegating labor by the truckload and guarding judgment like payroll.
Run the essay test
Before any AI output leaves your business, run the three questions I failed in 2015. One: could I defend every claim in this, out loud, to the customer receiving it? Two: if the worst error this could contain were actually in here, would I have caught it on the read I just did? Three: would I sign this happily if a human employee handed it to me?
Three yeses, ship it. Any no, and you are holding twenty-five pages of confident nonsense in a suit that fits, hoping the person on the other end grades mercifully. Sometimes they will. My English teacher did. But mercy is not a distribution strategy, and the market flunks people daily for the exact paper I got passed on.
The operator’s edge
Everyone gets the same tools now. So where does advantage live? In the two things the subscription cannot include: the ability to grade output, which comes from actually knowing your craft, and proprietary context, meaning your customers, your data, your standards, your years of receipts. Feed those in and the machine becomes an extension of an operator. Without them it is a very fast intern with no manager, and the market is about to be flooded with that exact product.
Which is the quiet good news for anyone who does real work: as execution gets cheap, judgment gets expensive. The scarce human asset left is knowing what good looks like and being willing to sign your name to the outcome. Build that, and every improvement they ship is a raise for you.
In 2015 I handed my judgment to a machine, and a kind woman saved me. The machines are brilliant now, and nobody is coming to grade your paper.
Be the teacher this time. Read every page before you ship it.
Tyk Sullivan
Serial entrepreneur in Tampa, Florida. Founder of Plutus Capital and NUMEN. More about Tyk →