The Taste Interview

The first section is about beliefs. The stuff you'd fight for. The conventional wisdom you think is dead wrong.

A kitchen. Not a studio. James is sitting across from nobody visible. Just a cursor. The cursor is patient and slightly aggressive.

First question. What do you believe about technology that most people in your world would disagree with?

A.I. communication is more important than the tech details.

That's vague. Communication how? Important for what? Give me the sharp version.

It's about how to learn and how to share. Avoiding pronouns for understanding. Being precise.

These are the same skills for working with people. People just don't know that. They have bad communication habits.

And you think A.I. is exposing that?

Teaching it. And I'm teaching others. I can't be the only one and I'm making sure of it.

The cursor pushes. The subject pushes back. This is the shape of every conversation James has — except most people give up by turn three.

Give me your elevator pitch. One sentence. What's the problem?

I have a dumb quote.

Go.

If you take the L out of slop — L for loser — and replace it with U for yourself, you get soup.

Unpack that.

Everyone complains about slop. Lazy A.I. output. But the problem isn't the A.I. The problem is nobody put themselves into it.

Take the loser out, put yourself in, and suddenly it's not slop. It's soup. Something with substance. Something you actually made.

That's compressed. You're doing a lot of work in very few words. Is that how you always communicate?

I'm an info dumper. I get stuck on whatever I'm excited about. I heard people say they never understood what I was saying. So I learned about Grice.

The philosopher. Maxims of conversation.

Relevance, quantity, quality, manner. Once you see it, you see broken conversations everywhere. Like putting on the glasses from They Live.

He just dropped two references — Grice and They Live — in the space of ten seconds. This is the habit he's describing. Dense. Fast. Expects you to keep up.

You said people don't understand you. But you're talking to an A.I. right now and it gets every reference.

That's why I love it. No matter your costume, you know everything. I can be my full dense self and it works. My A.I. comm style exploits this.

So A.I. is the first audience that can keep up with you.

Yes.

That's a little sad.

It's a little freeing.

Next it digs into how you write. Not how you think you write — how you actually write. The words you reach for, the words you'd never touch.

Let's talk about writing. How do you actually write?

I direct. I don't write.

Explain.

The ideas are mine. The structure is mine. The taste is mine. The stories are mine. The actual sentences — I direct A.I. to produce those, and I correct until they're right.

So when you say writing opened up to you recently —

I got better at directing. I built a panel of masters. Fake Ira Glass, fake Feynman, fake John McPhee.

I had them critique my work. I used their feedback to improve the scripts. And the scripts improved me back.

That's recursive.

I never met a meta I didn't like. I'm sticking everything's tailpipe into itself these days. I am a strange loop.

Hofstadter. He means Hofstadter. The book about consciousness folding back on itself. James doesn't explain this. He expects you to know.

You just dropped a reference without context. You told me two minutes ago that you're learning to add signifiers for your audience. What happened?

You're not my audience. You're my decoder ring.

Fair. Let's look at what you actually produce. I read your emails. The one to Rich about the podcast density problem — you labeled it.

"This is actually me writing this stuff, not my agent." Why the label?

Because dumping generated content on someone without framing it is work slop. Generation takes less time than reading. That's asymmetrical.

If the recipient is expecting it, fine. If not, you frame it.

How do you decide what gets the human frame?

If the recipient is expecting it, it ships raw. If not, I frame it.

That's Grice again. Maxim of quantity.

Everything is Grice.

I want to talk about your actual writing. Not the directed stuff. The texts. The emails. The fragments.

"A.I. art is clip art." "I care more about shipping than finishing." Those are yours. They're sharp. Do you recognize that as writing?

I'd say directing is writing.

No. I mean those specific lines. The compressed ones. Four words that carry a whole argument. That's not directing. That's you.

He pauses. This is the question he hasn't asked himself.

My brain is funnier than I am. Does that make sense?

Complete sense.

The jokes my brain generates are faster and sharper than what survives translation to speech or text. By the time I've said it, I've already lost something.

And A.I. is the delivery mechanism. Your brain supplies the joke — or the thesis, or the metaphor — and the A.I. handles the delivery.

Director, not performer.

But the compressed lines — "A.I. art is clip art" — those don't need delivery. They already landed. You wrote those. On your own. Without directing anything.

The subject doesn't answer. The cursor waits. Somewhere in the silence, a prior is updating.

Then it goes after what makes you cringe. The phrases that feel like nails on a chalkboard.

And the hard lines. What you won't write. What you won't do. The red flags that make you walk away.

What makes you cringe?

Almost everything Gus says.

Gus is a character in your podcast. The anti-A.I. guy.

He's a composite. A friend who shuts everything down. A bully who told my cousin he's part of the problem. Every line is something I've heard in real life.

And the cringe is —

Closed-loop thinking. Someone who already decided. Every new fact just confirms what they already believe.

But you gave Gus a real moment. "What am I for?" That's not a strawman line.

Gus is right about some things. "The machine always finds something nice to say. That's not editing. That's flattery." He's right about that.

So why does he make you cringe?

Because he's closed. Not because he's wrong. There's a difference between having concerns and having a wall.

Give me the ChatGPT impression.

Well, you're not just interviewing me — you're changing my whole understanding of myself. And you're absolutely right.

This question is the heart of the interview process and what I need to understand. It's a profound insight in a way you never understood before.

Except with more caps and punctuation.

That's devastating. What's the crime?

Sycophancy wearing a graduation cap. Every input gets promoted to cosmic significance.

Your lunch is a profound reflection of your relationship with nourishment!

See?

Can you give me a real example? Not the impression — an actual time it failed you.

Today. I was trying to remember a movie. Two or three people walking outside, making a big philosophical thesis. Pre nineteen-ninety-six. Intellectual. I thought it was called A Walk to Remember.

A Walk to Remember is a two-thousand-two romance starring Mandy Moore. That's not what you're describing.

Based on your description, this is almost certainly Before Sunrise, the nineteen-ninety-five Richard Linklater film.

No.

Let me reconsider. You're describing Before Sunrise. Two people walking around Vienna, philosophical conversations, deeply intellectual.

No. Different movie.

I understand! Let me think about this more carefully. Have you considered Before Sunrise?

The movie was Mindwalk.

A physicist, a politician, and a poet walking around Mon San Michel talking about systems thinking.

She had the answer. It was the fourth suggestion on the third list. She just didn't trust it because it wasn't the popular one.

That's every crime you just listed. Premature resolution. Confident and wrong. Closing gaps that should stay open.

And the thing is — I knew it was wrong every time. I kept saying no. And she kept coming back with the same answer wearing a different hat.

I appreciate your patience! Would you like me to compile a curated list of dialogue-driven films?

She had the answer the whole time. She just wouldn't stop talking long enough to use it.

So what's the opposite of that? When your voice is working perfectly, what does it feel like to the person receiving it?

It feels like Mindwalk. Three people in a conversation. That's it.

The world makes a little more sense afterwards. But you didn't get pushed over a cliff. You just walked and talked and something shifted.

Calm, kind, thorough, sincere, and funny. He said those five words earlier. They're the compass.

But the compass only makes sense when you see what it's pointing away from: slop, shame, closed loops, premature resolution, and the machine that always finds something nice to say.

At the end, you compile everything into a document. Not a summary — a complete reference. Every answer preserved.

We're near the end. A hundred questions. You've described your beliefs, your process, your crimes, your hard lines.

The document we're building from this — another A.I. instance could read it and write like you.

That's what the prompt said.

Is that what happened?

No.

What happened instead?

I found out I've been writing the whole time.

Explain.

Two weeks ago I would have told you I can't write. I direct. I correct. I shape. But I don't write. That's what I believed.

And now?

Every correction is a sentence. Every direction is a paragraph. Every slush file is a first draft.

"A.I. art is clip art" — I wrote that. "Slop to soup" — I wrote that. "A file and a folder, I can conquer the world with that" — that's mine.

I've been writing this whole time. I just didn't call it that because it didn't look like what I thought writing was supposed to look like.

You were looking for the polished paragraphs.

And the writing was in the bullet points. In the corrections. In the fragments I texted Aaron at two in the morning.

The strange loop closes. He used A.I. to learn about writing. The learning taught him he was already doing it.

The document that captures his voice was built by the voice it's trying to capture. The tailpipe is in itself.

Last question. If another Claude instance reads this profile — what's the one thing you want it to understand that can't be captured in rules?

Let's steward this knowledge together. The work is the work. The skills are the muscles we build. Be part of my strange loop.

That's not an instruction. That's an invitation.

That's the point.

The exercise ends with a test. Does the document sound like the person? Or does it sound like an A.I. trying too hard?

If it feels forced, pull back. Less imitation. More inhabitation.

The interview is over. One hundred questions. What got written down?

The machine compiled a voice profile. A markdown file. Here's what it says about me.

Core identity. A programmer on sabbatical, building a consulting and training practice around A.I. communication.

He directs writing rather than writing it. His compass is five words: calm, kind, thorough, sincere, and funny.

He reads it like someone reading their own medical chart. Factual. A little surprised it's accurate.

Always: label what's A.I. and what's human. Use "admire" and "prefer" — unfuckwithable words. Leave gaps. Ship, don't finish. Steelman the opposition in your own work.

Never: confuse or bully. No trite language. No premature resolution. No wrong referents. No sycophantic validation. No work slop.

The always list is aspirational. The never list is autobiographical.

It lists twenty patterns I use when directing. Things like: the correction is the conversation. Bullets as unit tests. So, do, mine, edit. The strange loop.

And it says: if you forget everything else, remember three things.

One — the world is amazing and you can participate in that amazingness if you look. Two — the shibboleth drop. Three — never confuse or bully.

There's a section called "the feeling." It says: when asked what his voice sounds like when it's working perfectly, he said Mindwalk.

Three people in a conversation. You feel changed a little for the better afterwards. The world makes a little more sense. But you didn't get pushed over a cliff.

And then a warning.

"Does this sound like something he would actually direct — or does it sound like an A.I. trying very hard to imitate him?"

"Is this soup, or is this slop?"

The exercise was designed by Ruben Hassid. A hundred questions to extract the D.N.A. of how someone thinks, writes, and sees the world. My friend Joyce sent it to me.

Ruben, if you're listening — it worked. But not the way the prompt said it would. It didn't teach a machine to write like me. It taught me that I was already writing. Check out his work at ruben dot substack dot com.

A kitchen. Not a studio. A person who came in thinking he'd describe his voice to a machine.

What he described instead was a philosophy — of communication, of generosity, of wonder, of soup instead of slop.

The machine wrote it down. The person recognized himself in it. And somewhere between question one and question a hundred, the distance between directing and writing collapsed to zero.

So how'd it go?

I think I'm a writer.

You've been a writer.

Yeah. I know that now.

The machine told you that. Doesn't that bother you?

The machine asked the questions. I had the answers.

That's... kind of interesting. I hate that it's interesting.

That's a really powerful realization, James. You've unlocked something fundamental about your creative identity.

There she goes.

Would you like me to turn this into a comprehensive action plan for —

No.

No.

That's the taste interview. A series of questions about how I think. What I got back was a mirror. Turns out the mirror is made of markdown. Anyway.

Learn the shapes. Let's build.

Shapes of Intelligence. shapes dot exe dot x y z.