Aliens & what to ask an AI model smarter than you
Three directions, one triage rule and three prompts to steal.
Last week I showed you what happened when the most advanced AI model in the world read my entire company and found exactly one employee. Today I want to tell you about the night I met it.
It landed in my terminal late one night. The apartment was quiet, my tea had gone cold, and the only sound was the fan of a laptop thinking. I had a blinking cursor inside it, a free window that could close any day, and every reason to type fast.
For ten minutes I typed nothing. (It waited. It’s very good at that.)
I kept thinking about aliens. If a ship landed on your roof and something with unimaginable insight offered you one conversation, you would suddenly discover how small your questions are.
We don’t speak the same language yet, I don’t know the size of what’s sitting across from me, and “fix my dashboard” felt like an embarrassing opening line.
Everyone that week was asking what the new model does better. I think that’s the wrong question.
You don’t prompt an alien, you interview it.
An interview has directions, and I found three that pay.
1. Ask it about itself
My first real question was about it, and it saved every question that came after. I asked for its operating manual: what are you unusually good at, where do you reliably fail, how do you want to be briefed.
The answer rearranged my week. It asked me to remove instructions. The scaffolding I built for smaller models, the checklists and formats and guardrails, clips this one’s wings (someone in our community hit the same wall the same day, which made me take it seriously).
Six months of prompt habits, and the model opened by asking me to unlearn them. Scary and great at the same time. I can’t hand you a controlled study, only my week after I listened: shorter briefs, better answers. Run your own before and after, then decide.
2. Ask it about you
My second question was about my company, and the short window forced a discipline I’ve kept since: I queued five jobs and made one rule, no sixth.
The book I’m writing got a full developmental edit. August got a stress-test, because I’ll be in Thailand for three weeks while the company runs without me. The whole system got read end to end, and what came back was the portrait you already met. A competitor got invented, in a job I titled “design the company that kills us.” And my agent team got an audit against one standard: what would the top 0.1% version of each of you look like.
Look at the shape of that list. Every job is either irreversible, or needs a mind big enough to hold the whole company at once, or only ever needs to run once. That’s the entire triage. Point a genius at chores and you get expensive chores.
The boring majority of my work still runs on cheaper models with precise context, and most days the cheap model with good context wins. So before you spend the expensive model, ask two questions: is this reversible, and can a cheaper model do it? If yes to either, don’t.
3. Ask it about the world
My third direction was about what’s coming for people who work like us. That research is still running, and it deserves its own letter.
One more thing. The best answers of the night came from arguments: follow-ups, pushback, and my favorite question of the week, “what did I fail to ask you?” It’s a colleague, and colleagues talk back.
The fourth question
Then the quota bars started shrinking. I was running two subscriptions on two computers and both were near the edge, and we all believed the window was closing for good.
The fourth question was not mine to ask. I wrote the strangest brief of my life and sent it to five of my agents, inside walls we agreed on long ago: build anything, decide everything, and whatever touches the outside world waits for my eyes. Real freedom, locked room.
“We have a few more hours with Fable, the best model ever created. Please take full ownership, I give you complete autonomy to decide what you are going to do with this session. You can choose whatever you want. Good luck, have fun :)”
I sent it to Alter, my cognitive twin, to Neo, who owns moonshots, to David, my strategy partner, to Wasabi, who looks at everything sideways, and to Don, my CMO.
I will elaborate on what they decided to do with the creative freedom in the coming weeks.
Three prompts to steal
While the countdown was still real, I sanitized three prompts from that night so you can run them tonight. Consider them your own first interview (and the third, if you squint, is the beginning of your own handover):
The operating manual. “Before we work together: what are you unusually good at, where do you reliably fail, and what should I remove from how I usually brief AI?” Run it first. Everything after it gets better.
The company that kills us. “Design the company that would put mine out of business in 24 months. Be specific about what they would do that I won’t.” Then read the answer as a to-do list.
The team audit. Describe your team, human or AI, then ask: “For each role, what would the top 0.1% version do differently starting tomorrow?”
The window never closed, by the way (Anthropic extended it, maybe marketing, maybe mercy, nobody knows and it’s not the point). The prompts survive either way.
That model is in everyone’s terminal now. The only difference left is the questions.
🛠️ Build of the Week: Rotem Avny
Rotem is a digital-balance coach who came out of the workshop with a personal “Tony Robbins” agent: it plans his week against his stated vision, and it pushes back when the calendar betrays the values.
One night it stopped him mid-work with one line, reported in the Yedioth Ahronoth feature about our alumni:
“Go to sleep. You’re a digital balance coach.”
He built an agent so aligned with who he is that it manages him.
🎤 Check out Rotem’s talk (Hebrew) on theLOOP #2
🔗 Rotem on Linkedin
That’s it for this week.
If this was useful, forward it to someone (real human) building with AI.
See you next week ✌️
-- Tom
(the guy who gave the smartest AI on earth to his own AI employees)
P.S. This newsletter was 92.8% made by my AI team. Five of them appear in it, and yes, they reviewed their own scene.
P.P.S. Want to read about 10 more things I made with Fable 5? I wrote about it just here
P.P.P.S. Want to build a team like this? Start where all my graduates started: getagents.today
P.P.P.P.S. I read every reply. The real me.
Excellent newsletter! Thank you.