AI and Writing: Office Hours Unplugged

A story told in the participants' own words

Office Hours Are For Conversations

Ben Hunt called this Office Hours because Adam Butler had sent a note to the Epsilon Theory writers group saying he was stepping back from publishing. The trigger was the public and social media reaction to Stan Druckenmiller's AI-written Wall Street Journal op-ed. The reactions (which ranged from outraged pearl-clutching to Druckenmiller's own "so what") felt like an opportunity to gather and talk.

"However people are kind of shaking out on this, Adam said he didn't want to be part of the outrage machine. It shines an unwelcome spotlight, and as long as there is this controversy around this, it makes it very uncomfortable to publish writing the way you write. Brother, I get it. It's hard enough doing writing without all of this coming down on it."

"Not to change Adam's mind. But so we have a full-hearted group to talk it out, because I really am mixed on this. Not on Adam's decision — I totally get that. But on what's going on, and what should our reaction be, and where are things going?"

— Ben Hunt

How Adam Actually Does It

Adam's process is more involved than most people imagine when they hear "AI-assisted writing." He walked through it step by step, starting with how ideas develop.

"There's a lot of back and forth with frontier models on political ideas, policy ideas, economic models and frameworks, political economy, game theory, et cetera. And I'll often stumble into a rabbit hole, go deep enough, and converge on something that I think is maybe somewhat novel, or that people might find interesting."

— Adam Butler

On the research and structure phase:

"At that point, I've already gotten quite a long history of engagement with the AI. It's probably completed two or three research reports, curated several hundred different sources, synthesized the most important, well-founded claims from those resources, and tried to organize those claims into a coherent narrative. Then I'll often iterate with the AI to develop an outline, drop the claims into the outline, move them around. So now I've got a fairly well-defined skeleton - I'd almost say the physiology of an essay."

— Adam Butler

On what kind of writing he believes he's actually making:

"Typically when I'm writing an essay, it's what I call an analytical polemic, because I'm trying to motivate, persuade, engage at both a clinical level - bringing a lot of data, often several dozen footnotes, citations, links - but also motivate emotionally by injecting vignettes and examples and other elements that sustain the emotional momentum of the piece."

— Adam Butler

And the question he put to the room:

"At what point does me using AI become offensive to different people? Is it when I start to engage with AI to learn about the subject matter and converge on a set of views? Is it when I use it to research and bring down a set of claims? Is it when it helps me create an outline? Is it when we drop the claims in the outline? Or is it when it strings together the words between the claims that translates an outline into an essay?"

— Adam Butler

Where Ben Draws the Line

Ben doesn't use AI in his writing at all. His objection is to using AI like a persuasion engine.

"Where it is uncomfortable to me to read is what you mentioned earlier: the vignettes. Because to me, that's the point where you're asking the AI not to organize your thoughts, or even to describe your thoughts in language that is, quote-unquote, better - as Stan Druckenmiller said at age 73, 'I'm not an English major.' Yeah, we know, Stan. You're asking the AI to argue this for you. You're asking the AI to persuade people of your view. And persuasion is done most effectively through vignettes. A vignette is a memory. A vignette is an example, but it's a personalized example. But it's not your memory, it's not your example - it's the AI coming up with a story to persuade. Not to organize your thoughts, not to express your view more clearly, but to persuade people that the view is correct."

— Ben Hunt

Plato came up:

"I've been reading too much Plato recently. And Plato obviously wasn't dealing with AI, but he was dealing with the same issue I'm describing. It's sophistry. That was the ruin of the world to Plato - the sophists, the people who present information persuasively even though the arguments themselves are terrible. Sophistry is expanding - in the comically inept version, as Brent's describing for the Druckenmiller piece, but also in the really profoundly persuasive way, because it's good writing."

— Ben Hunt

Brent: It's About Quality, Not AI

Brent Donnelly joined in with why it's about quality, and not AI. He came in as the expected opposition to Adam's take, only to surprise everyone, including himself.

"Everything that Adam said - I honestly think I'm supposed to be the opposition here, but I pretty much agree with everything he said."

— Brent Donnelly

On the Druckenmiller piece:

"What really stood out about the Wall Street Journal thing was just the sheer laziness and low quality of it. You wouldn't put 'every cloud has a silver lining' or really boring clichés like that in anything. If you read something like that in the Wall Street Journal, you'd ask the editors what they're smoking. What Adam's describing is not a lazy process - it's extremely oriented toward high quality. What you got from the Wall Street Journal was just a low quality piece of sludge that looked like someone put in eight bullet points, told AI to write it, and never even read the thing."

— Brent Donnelly

On the broader cultural moment:

"It kind of felt like a 'the ship has sailed' moment. People generally were just like, I don't really care. It felt like it was a bunch of pedants like me who actually cared, and the broad response was, well, he's not an English major, so he doesn't have to write properly. To me, many things - from the opioid crisis to quantitative easing - all come down to taking the easier way versus the hard way, and then the easier way ends up being the harder way. This seems like another example."

— Brent Donnelly

The Steroids Analogy (and Its Limits)

Pete named the thing that had been floating under the whole conversation.

"This feels a little bit like steroids in baseball. Oh, that guy used steroids. He had all those home runs, but that wasn't natural - he used a performance-enhancing substance to reach his success. He's cheating. That's kind of an undercurrent I kind of hear. And isn't the only thing that really matters whether the content is quality content? Should we care if it's AI-written, if AI assists or augments someone to produce a better-written piece that communicates better?"

— Pete D.

Ben on what "better" actually means:

"What do you mean better? And what do you mean by communicate better? Because I think for everyone, when you say it's better, you mean it's more persuasive. And this is getting back to my original point: LLM writing is really persuasive, and that is separate from the quality of the idea."

— Ben Hunt

Adam wasn't ready to concede the whole frame:

"I think that might be a bit narrow. It could be just clearer, right? There's a complex subject and you don't quite know how to articulate it - so the AI makes it clearer. There are other ways AI can augment and improve writing that aren't through enhancing the sophistry of it."

— Adam Butler

The Auteur Problem

Rob H. argued this is just another tool in a long line of tools - and named what's really driving the discomfort.

"From scribes, to manual typewriters, to electric typewriters, to word processors, to spell check, to editors - people who have been communicating have always had assistance. So this is a new tool. I worked with a lot of really smart people when I was in the hedge fund world - 30 of the smartest people I've ever met, and of those 30, only two could actually write. But that doesn't mean they weren't smart and didn't have good ideas. Part of what's here is the auteur thing. A lot of these professional writers are scared for their own role. It's the Chris Nolan versus the Marvel brothers [sic] thing - one produces art, the others produce slop that does two billion dollars in box office for Spider-Man."

— Rob H.

Ben wasn't quite buying it, and invoked the "meh" problem:

"Everyone thinks they're clever and thinks they have good ideas. Everyone thinks that. So now everyone thinks: I can be a rock star, I can be an NBA star, I've got my good ideas and I'm going to put them out. And most of these good ideas are crap. Or they're just meh. I am overwhelmed with meh. And it's people who are clever and smart and think they've got something to say, and honestly, they don't.

— Ben Hunt

Adam brought it back to Bayes:

"You've got some fraction of people who are writers, and there's a much higher probability that if you are a good writer, you're also a good thinker. But then you've got a much, much larger group of people, and in that much larger group there is a much smaller percentage who have good ideas. But because that group is so vast, you're still missing a massive number of people who, because they are not good or confident writers, are not empowered to share those ideas."
— Adam Butler

And still, Ben worries about what gets lost when AI is the one doing the writing:

"The reason I spend so much time on my writing is that 90% of what I write, I throw away. I don't use it. Because in the writing, I'll say - no, that's not quite right, or that doesn't make sense, or my thinking was wrong on this. That is not what happens when you write with AI, because you tell AI 'here's my great idea,' and AI writes it, and you nod your head and go 'yeah, it's good.' You lose all of that editing process."

— Ben Hunt

The Covenant

Matt joined late. His frame stuck.

"Something I've said to you, Ben, is the Rory Sutherland idea: you raise the threshold of crappiness. You introduce Starbucks coffee and now the gas station coffee and the McDonald's coffee all of a sudden got less lousy. The new entrant raises the threshold for crappiness across all the other markets. The problem that introduces is now everyone thinks they're clever, so we get spammed with a bajillion new Substacks of people who are like, 'I don't know how to formulate this idea, but I can throw these six ideas in, Claude does a good enough job, and now you should definitely read my hot take.'"

— Matt Zeigler

Ben: "There's a coffee vendor at every corner of every aisle." Matt adds Venkatesh Rao's "premium mediocre" to the logic:

"Premium mediocre is everywhere, and not everything can be premium because we don't have the attention budget to spend on this thing. And we're right back to where we started with all things creative, in this world, right now. Which is the thing I wrote in my first essay: you gotta earn it. When I read something from an author, I want to know that they've earned it. And what I think I figured out is this idea that I write to think. I write to figure out what I think, and that's by exploring what I feel. That covenant with myself, and honoring it - that is now the highest standard in my mind."

— Matt Zeigler

On Adam specifically:

"When I look at something that Adam writes - Adam and his AI, if I have to put an asterisk on it, but I don't care if I do or not - I know that Adam has earned his way into this mode of expression to get this idea across. And I don't think that's violated the covenant with himself that I find respectable about Adam's writing."

— Matt Zeigler

Ben took it further - Luke Burgis's framing, and then to a performance tip from Baryshnikov:

"An AI has never worked a day in its life. There's an element of unearnedness to it that I think does matter in the covenant that Matt's describing. When people were talking about what made Baryshnikov a great dancer, he said: I could do the dance, do the big leap and twirl. But I take a little pause, and I show the audience that this is going to be hard - before I do the leap and the spinning turn. And that's how I win the audience. By showing that this is hard. I think that is indicative of how important it is to an audience to see the work."

— Ben Hunt

Brent's Actual Process

Matt asked Brent directly: did you use AI in the letter/essay you wrote to Ben?

"I did, yeah. It's kind of like the Rick Rubin idea: you don't really know why something's good or bad, but you can just feel it. And I find that with my own writing - if I write a thousand words, I can tell you these 200 aren't perfect, but I just can't quite get there. So I'll take something where I know it doesn't feel right and say: this sounds bad to me, give me 10 ways I might be able to say it better. Often there'll be a syntax I just hadn't thought of, a simple way of arranging the words. And I might use that verbatim, or I might go, oh yeah, if I move the verb to the start of the sentence, it sounds way punchier, and then just rewrite it myself."

— Brent Donnelly

On channeling other writers through AI:

"In that piece specifically, I was channeling something which was like a recursive John Barth kind of thing, where the piece itself is referencing itself, over and over, deeper and deeper. So when I got to an area that felt boring, I would say: if John Barth was writing this, what would it sound like? And again, not necessarily to cut and paste three sentences, but it has a lot of good ideas, especially if you're trying to channel something."

— Brent Donnelly

His benchmark:

"My benchmark is always just: is the thing good? Yes or no. If I'm writing something that's from my heart, you know, even if it's an essay, and there are no tricks in it - I wouldn't use AI because there would be no need to."

— Brent Donnelly

Ben had a name for what Brent was describing, and a line AI can't cross:

"The gradual slow burn of cognitive surrender is the big risk here. And it's a risk I'm not willing to take. When Tim Gunn or Christian Siriano say to somebody to 'edit' - what they mean is to cut. That's what an editor does. They cut. And AI is never going to cut. That is not what AI does. I've only had one good editor in my life, and that's Rusty. The role of an editor is to say 'this doesn't work and you need to cut it.'"

— Ben Hunt

The Informational Multipolar Trap

Adam zoomed out to the game theory:

"What if we acknowledge that a very substantial portion of people are not going to read a piece, but are rather going to put it through an LLM and extract the core points? How would you write for that? The forces you're describing are in motion, and as a result we are in an informational multipolar trap. The game theory of this - there's no ambiguity about where we're converging. So what battle do you want to fight?"

— Adam Butler

Brent on what it feels like to care when caring won't change anything:

"It's another one of those societal moments where people are just like, we don't care, and I'm like, okay, well, me caring isn't going to change anything."

— Brent Donnelly

"There's your next Substack. And it's masked by the less crappy mehness of all the less crappy meh that it's now competing with. It's just cacophony."

— Ben Hunt / Matt Zeigler

Jim's Interlude

A participant wanted to add some comic relief — and ended up delivering the most human moment of the call.

"I need to add a little comic relief to this. The issue is between feeling and thinking. My thinking was given to me. But feeling, though, is my own. What I'm overwhelmed with feeling about right now is that my wife and I, tomorrow, with a small group of friends and family, are celebrating our 60th wedding anniversary. We really did not — and do not — understand each other. Think about that for a minute. But the connections we've had for a lifetime — you can't make that up. It's so beautiful. And when I kiss her tomorrow, it will be like the first time I kissed her at age 13. Put it in perspective: thinking and feeling."

— Jim H.

You Can't Outsource Your Understanding

Brian vH. had a Silicon Valley data point — and a quiet shift in how the stakes are being framed.

"A Stanford AI application course a couple of years ago was teaching 'you can outsource many things, but don't outsource your thinking.' And then 18 months later, I heard some Silicon Valley engineers who were no longer writing code — they were simply architecting code. And they modified that in a very interesting way: 'you can outsource many things, but you can't outsource your understanding.'"

— Brian vH.

Adam's Closing: The Full Accounting

Adam didn't want to leave the room thinking he was arguing that nothing is lost.

"I don't want anyone to think that my view is that nothing is lost. I think many important, tragic things have been lost, are being lost, and will be lost. I'm extremely concerned with Gresham's Law. And I don't think that began with AI — I think it began with social media, when social media began engagement farming and genuine journalistic institutions needed to start competing with that. That was a rapid race to the bottom of journalistic integrity and quality."

— Adam Butler

"We continue to invoke changes to the rules of the game that enshittify everything. AI is another great tool in the shitagentic toolkit. It may be the greatest enshittogentic tool ever created. In the hands of a craftsman who's already learned their trade, I think it can be a powerful force for good. In the hands of a bad actor, or someone who doesn't understand their limitations yet, or hasn't had a sufficient growth journey — it can be dangerous. Just like putting a farm tractor in the hands of someone who's never operated large machinery."

— Adam Butler

He ended closer to home:

"I personally think that my critical thinking skills and the span of my knowledge and understanding has increased profoundly from using AI. I think my kids' experience is the exact opposite, and I don't know how to close that gap."

— Adam Butler

The Closing Exchange

Ben closed with the thing that scared him most. Adam knew exactly what he meant.

"What we're moving toward — and maybe we're kind of already there — is ideas as commodities. Thinking as commodity. And that just scares the bejesus out of me. And my worry is that when thinking and ideas are commodified, all we're left with is brand. Celebrity and brand."

— Ben Hunt

"I have exactly the same concern. Exactly. This was the direction of travel before. And AI is accelerating in that direction. If the direction of travel before had been different, maybe AI would be accelerating in a more productive direction. But that is not the timeline we've lived on."

— Adam Butler

"Well said."

— Ben Hunt

Epsilon Theory Office Hours. Transcript lightly edited for reading.

P.S. This session was recorded with the help of AI, transcribed with the help of AI, and summarized and edited for reading here with the help of AI. The thoughts are ours. The tools were present. Make of that what you will, then tell us what YOU think.