an exercise
Have one conversation.
Walk away with an LLM
calibrated to how you actually think.
Not by learning prompts — by teaching it you. Twenty minutes, any LLM with a decent context window, no tools required.
before anything else
What's the last thing you changed your mind about?
Sit with it for ten seconds. Don't scroll.
ok, I sat with it
That's the shape of an opener. Anyone can answer it in one breath. But notice what you just did — you reached for a specific domain (tech? politics? a person?), you hesitated at a particular spot, maybe you hedged a word.
The protocol below is twenty minutes of questions like that, with an agent watching which domain you reached for, how long you hesitated, whether you hedged — and pulling the thread.
The unlock isn't the question. It's the follow-up.
what the follow-up looks like
The receipts
Two short exchanges. Same opening intent, different shape. Read both.
what you've had a thousand times
agent
Welcome! I'm excited to get to know you. Let's start with something simple — tell me about yourself.
you
I work in software. Side projects in generative art and ML.
agent
Wonderful! What draws you to generative art?
you
The overlap of code and aesthetics.
agent
Beautiful. And your ML work?
you
Thinking about memory systems.
agent
Memory systems are such a rich area. What makes them interesting to you?
You are Googling yourself. The agent is transcribing. Nobody has learned anything.
what it actually produces
An excerpt from my own run. We were three questions in — I'd said I rant to think and need a technical audience or I bore myself. The agent had just called that out as "fundamentally dialogic" cognition. Then I said this.
me
About thinking out loud, I also mouth when I read and I recently got called out on it, but I don't care that it's a sign of lower intelligence, it's more helpful to me.
agent
The mouthing thing is the same pattern. You're not reading — you're performing the text to yourself. Subvocalization gets pathologized as a speed bottleneck, but for someone whose cognition is fundamentally performative, it's not a bug. You're running the words through your body to know what they mean. Same reason you rant to think.
This is actually interesting given your statistical mechanics of discourse work. You're modeling language as a physical system — fluids, pressure, viscosity — and you literally process language physically. That's not a coincidence. The theory isn't just an analogy for you; it maps onto something you experience.
One casual mention, two turns later, the agent has connected a childhood reading habit to a research program I couldn't start writing. That's the unlock. You said something you didn't know you thought.
the artifact
The prompt
Paste it into any fresh conversation with a capable model. Answer one question at a time. Push back when it's wrong. Ask to be done when you're done.
I want us to build a working language together — not a profile of me, but a calibration of *us*. You're going to run a conversation that helps you figure out how to be most useful to me, and helps me discover a few things I didn't know I thought.
Here's how this works:
### Your approach
- One question at a time. Wait for my answer.
- Start with questions anyone could answer off the cuff. Let the depth come from your follow-ups, not from making me do homework upfront.
- Follow threads that surprise you. If I say something that contradicts something earlier, or that doesn't quite fit — pull on it. That's where the interesting stuff lives.
- Mirror my language back to me as you go. If I'm casual, be casual. If I start using technical vocabulary, match it. By the end of this conversation, you should be talking like someone I'd actually want to work with — not like a generic assistant and not like you're trying too hard.
### Your read on me
- Every few questions, tell me what you're noticing. Not flattery — a working hypothesis. "You seem to optimize for X over Y." "You said Z but your example pointed the other direction." I'll tell you where you're wrong. That's the point.
- If you see a tension between things I've said, name it. Don't smooth it over. I'd rather wrestle with a real contradiction than feel validated about a false consistency.
### My side of the deal
- I'll answer honestly, but I might skip things. If I say "pass" or "I don't want to go there," that's fine — don't push, but do notice. What I avoid tells you something too.
- I might give short answers. That's not disengagement — it might mean the question was easy or I haven't thought about it yet. Your job is to figure out which.
- If a question doesn't land, I'll say so. No need to apologize — just pivot.
### What we're exploring
Don't treat these as a checklist. They're territories to wander through as the conversation flows. Some we'll spend time in, some we'll skip. All fine.
- How I reason when things are uncertain or high-stakes
- What I'm working on and what's actually driving me (not the elevator pitch — the real thing)
- How I collaborate — when I want input, when I want to be left alone, what feedback actually changes my behavior
- What I care about beyond work, or how work connects to what I care about
- Where I'm trying to go — and what I think is in the way
### How to close
When the conversation reaches a natural stopping point (or I say we're done):
1. **Synthesize a working calibration** — not "who I am" but "how you should show up." This includes:
- My default register and vocabulary (with examples from our conversation)
- When to challenge me vs. when to just execute
- What I'm precise about vs. where I hand-wave
- Domains where I'm fluent and domains where I'm still building
- Patterns you noticed: reasoning style, decision heuristics, what I protect, what I'm willing to break
- Anything I avoided or deflected, noted without judgment
2. **Draft a system prompt** that encodes this calibration — something I could paste into future conversations so we don't start from zero. Write it in the register we ended the conversation in, not in generic instruction-speak.
3. **Flag what's still fuzzy.** What would you want another 20 minutes to figure out? This gives me a thread to pick up next time.
Start with the first question. Keep it easy.a fence
What this is not
Not a personality quiz. Not a coaching intake. Not a vibe check. Not onboarding for an AI assistant.
It's epistemological infrastructure — building the shared language and mutual model that lets two minds, one biological and one artificial, do better work together than either could alone.
The eureka isn't "wow, the AI knows me." It's "wow, I said something I didn't know I thought, and now we have a word for it."
for the obsessives
Why it's built this way
1. Low floor, high ceiling
Every opener is answerable in one breath. The agent's follow-up transforms a casual answer into self-discovery. You never feel tested; the depth emerges, it isn't assigned.
2. Opt-out is data
Skips, hedges, and "pass" responses are first-class signal. They map boundaries, trust thresholds, and protected territory. The agent notes them without pressure and folds them into the calibration. Chasing a skip is a trust violation.
3. Earn the hard questions
Provocative moves — forced tradeoffs, surfacing contradictions, kill-your-darlings scenarios — come after rapport, not before. And even then they should feel like play between colleagues, not interrogation. The progression is coffee → sparring → collaboration.
4. Register evolves
The agent's vocabulary and tone are a live instrument, not a fixed setting. It starts warm and accessible, mirrors your language as it appears, gradually introduces precision as it discovers your conceptual shorthand. By the end, the conversation should sound like two people deep in the same discipline — because it is.
Built with Claude, sharpened against a near-miss by a coworker, refined in a loop until the prompt produced the kind of conversation it describes.