Most people use AI to get things done. You hand it a task — write this code, scrape this data, draft this doc — and it hands back an artifact. This is real, and it is useful, and I do it all day.
It also has a ceiling. The moment AI produces something past the edge of what you actually understand, you lose the one thing that matters: judgment. You can’t tell whether the result is right, whether it’s safe to ship, whether you’d stake a business decision on it. The artifact looks finished. You just can’t vouch for it.
There’s a second way to use the same tool, and almost nobody reaches for it. Instead of asking AI to produce the work, you ask it to produce a sharper version of you. The output isn’t a deliverable. The output is that you walk away better than you walked in.
The difference is in what’s left over. When AI does the work, you keep the artifact and stay exactly who you were — sometimes a little weaker, a little more dependent on the next prompt. When AI trains you, it leaves nothing behind but the thing it changed: you. And because you’re the one being sharpened, there’s no trust problem to solve. You’re not shipping its output. You’re shipping yourself.
The interrogator
Before an interview, I don’t spend the night reading about the company. I have AI interview me.
I tell it to play a hostile, impatient interviewer. Ask the hard version of every question. When I give a clean answer, follow up. When I wave my hands, call it out. Keep going until I can answer everything without stumbling.
The first pass is humbling. I think I know my own résumé, and then AI asks why I chose that method over the obvious alternative, and I realize I’ve been carrying a story I can’t actually defend. So I fix the story. I go again. By the third or fourth round, the answers come out clean.
Then the real interview feels easy, because nothing in the room is harder than what I already sat through alone.
Pressure-test before you ship
I do the same thing before any readout. Before the deck goes to a VP, it goes to AI first, and I tell it to tear the argument apart.
The logic is simple: if AI can’t follow my reasoning, the room probably won’t either. A question AI can’t resolve from my own slides is a question someone will ask out loud — except out loud, it costs me. Better to find the hole in private.
It does two jobs at once. It surfaces the obvious attacks — the thin sample, the confound I waved past, the chart that says less than I claimed — so I can have an answer ready or cut the claim before anyone sees it. And it forces me to re-read my own work the way an adversary would, which is when the small stuff falls out: the number that doesn’t reconcile, the step that doesn’t follow, the conclusion that outran the evidence.
Rehearse the room
The third move is the one I’ve come to value most. AI can stand up a rough version of almost any situation in seconds, and then you can live in it until it stops feeling new.
A negotiation you’ve never run. A hard conversation you’ve been dreading. The first time you’ll chair a meeting full of people more senior than you. You describe the setup, you hand the other side to AI, and you run it — badly, the first time. Then again. You find where it goes sideways, what they’ll push on, how the thing actually flows, all before it counts.
Most of the fear in a new situation isn’t the situation. It’s the uncertainty — not knowing the shape of it yet. Rehearsal doesn’t lower the stakes, but it removes the strangeness. By the time it’s real, you’ve already been here. You step in already in character, instead of spending the first ten minutes finding your feet.
Which direction the tool points
Notice what these have in common. The first way of using AI keeps you parked inside your current edge, and punishes you the moment the work drifts past it. This second way pushes the edge outward. Same tool, opposite direction.
And it compounds. The artifact AI generates is worth something once. The reps you put in against it — the interview you can now ace, the argument you can now defend, the room that no longer rattles you — those stay. AI logs off. The capability doesn’t.
None of this shows up in a demo. There’s no output to screenshot, nothing to ship when the session ends. That’s probably why so few people bother. But the work you make with AI is never better than the person directing it. So take some of the time the machine just saved you, turn it around, and point it back at yourself.