Evil Thoughts From Good Men
Halloween Edition: the true story of how a demonic typewriter found me
Architect-turned-writer, founder of Essay Architecture. Building pattern languages, software, anthologies, and community for essayists.
Architect-turned-writer, founder of Essay Architecture. Building pattern languages, software, anthologies, and community for essayists.
Essay Club ↗
Architect-turned-writer, founder of Essay Architecture. Building pattern languages, software, anthologies, and community for essayists.
Architect-turned-writer, founder of Essay Architecture. Building pattern languages, software, anthologies, and community for essayists.
Essay Club ↗
Architect-turned-writer, founder of Essay Architecture. Building pattern languages, software, anthologies, and community for essayists.
Halloween Edition: the true story of how a demonic typewriter found me
My doubt comes for Margin Muse (my editing app) comes from a few places, but the big one probably stems from the fact that people don’t want to pay for software subscriptions any more. Now they pay $20-200/month for Claude/ChatGPT, and that does everything for them.
Claude even added a comment feature to make it like Google Docs, which is basically the core premise of Margin Muse. Of course that won’t be as specialized as my own tool, but maybe it’s good enough for most people?
I’m also realizing that I can build my own interfaces directly inside of Claude. ie: You could type /essay-architecture in Claude and then get the radar graph with all 27 patterns scored. Or you could type /margin-muse and see a near identical interface to my own product. It's wild stuff.
Right now I think I have two kinds of users, writers and AI-nerds, and they want slightly different things. The writers barely explore Claude (so installing MCPs might seem odd), and the AI-nerds won't go outside of Claude. I suppose I can just focus on the writers, but I worry that software is a losing battle in general.
This is all complicated by some friends who actually seem to genuinely like the tool and want me to keep improving it.
Maybe I just need to focus on Essay Club, the in-person component, and operate under the ethos of building this tool mostly for myself as a personal power tool, and inviting friends to join, and it’s okay if most writers end up just using Claude.
My X algorithm likes to show me people dying: a cross falls on a man, a man falls from the catwalk at a football game, a cheating boyfriend jumps off a balcony, a drunk sorority girl falls off a balcony, a freak car accident that looks impossible to survive, a mother calmly calls 9-11 after she murders her whole family in a frenzy. I suppose if 121 people die every minute, then social media is just making the enterprise more visible.
Does the algorithm push this on everyone? Or, more specifically, does it have the data to prove that I personally like watching people die? I certainly don't enjoy, condone, or seek it. But if it's presented to me, it's sort of hard not to look (in the same way it's hard to drive by a traffic accident without a curious glance). Death was once rare to see, but now it's common, that is, if you see it once and let it loop from a place of shock.
And so you could blame me for looking or blame the platform for showing. Of course death goes viral if unregulated. Should it be regulated? I remember when you had to go to underground websites to watch people die. FWIW, I never submitted myself to watch Al Qaeda beheadings. Perhaps that's proof that I know not to intentionally submit myself to it. The difference here is that it's all just mixed in: a new product release, a life update, a motivational quote, a man getting mauled by a tiger, an SNL skit, etc., on it goes.
I imagine it could be simultaneously better and worse.
Better in the sense that we might realize the personal computing revolution, meaning the average person will be able to spawn their own software, networks, and mission-driven companies to make their lives better in meaningful and idiosyncratic ways. Much of the anti-tech rhetoric comes from an adversarial relationship where users have no control over what they use, and that’s quickly changing.
Worse in the sense that the rate of progress will be disorienting. Mathpocalypse but for everyone else. For example, I’m open but reluctant to the heretical idea that AI writing could get quite good. Many people base their assumptions on AI writing from the technology 3 years ago, which was exclusively slop, but in 3 years from now it might be producing stuff that's awe-inducing to read, undeniably good. This will be confusing, especially for writers. It helps to anchor internally (to write essays for yourself), but it’s hard to imagine the arrival of an alien-prosed mecha-soul not being slightly distracting.
To summarize, we might find ourselves intrinsically empowered but extrinsically disempowered. The barrier between imagination and reality is going to zero, but that’s going to cause a collective ego crash.
Also worth considering the possibility of a black swan. Of all the doom scenarios, I think the realistic one to brace for is something that’s equally as disruptive as the pandemic we recently went through. Except I think it'll be a cyber pandemic, where a lot of the characteristics are reversed. Instead of the virus being outside and everybody retreating onto the internet, the virus will be on the internet, and everybody will have to go outside. It's hard to know the degree of severity, but I imagine large swaths of Internet being down for extended periods of time, which doesn’t sound too extreme, but would have a lot of second-order effects.
So basically I think it will be easy to look around and say things are worse (like today), but hopefully possible for someone with strong purpose and principles to live better.
What’s the mission of this project?
A culture needs a way to surface its best essays each year. They help us make sense of the times, and they inspire people to write. Our current institutions aren’t doing this well. Literary curators are either fixated on a particular theme/aesthetic over timeless principles of quality, and social media algorithms reward whatever drives clicks and revenue. Great independent writers get overlooked. The Best Internet Essays uses agents to find them and curator a longlist for human judges, who decide the winners that get published in an annual print anthology.
How are publications discovered?
Publications are surfaced both through AI agents and human submissions. We’re building about a dozen agents, each with their own search heuristic to discover RSS feeds. From then on, every new post enters our database for review automatically. Once a publication is found, its essays are tracked for the rest of the year, and the writer doesn’t need to do anything.
Who picks the winners?
People do (the final decision is made by human judges). Agents handle the early stages only because no human team could read everything published online by independent writers. Each piece goes through three phrases. First, a quick check confirms it’s a full essay and not a news article or paywalled post. Second, a fast, low-cost scan scores it against the Essay Architecture pattern framework to analyze essay composition. Third, a frontier model evaluates the highest scoring ones with more accuracy. Michael Dean then reviews the longlist and narrows it to a shortlist for human judges, who make the call based on taste.
Where are the results shared?
Scouting runs through December 31, 2026. In January, the judges begin reading the shortlist, which gives them time to consider each finalist carefully. The winning essays will be published in a print anthology, scheduled for release in spring 2027. Each year's collection is meant to capture the best essays published online that year.
What data is stored, and how is it used?
We store only information that’s already public: each essay’s URL, title, author, publication date, and text. The text is used only to evaluate whether the essay is a contender for the anthology. Nothing we collect is used to train AI models. Writers whose essays appear in the anthology keep full rights to their work. Our goal is to find and celebrate great human writing, not to extract value from it.
Vacations with a baby are fun and obviously special, but definitely not relaxing, rejuvenating, or clarifying. After ten days in Florida, I come home to scrambled work projects, with my creative energy basically depleted. It wasn't until my first long walk on my second day back—which came with a climate shift too, from swamp air to New York fall—that I actually got the "reset" I typically expect from vacation.
The clarifying thought was "I want to live a principled life," and I still need to make sense of exactly what this means and how to do it. We probably all like to believe that we harbor good virtues, but we don't necessarily have a practice to stop and define them explicitly, or to coin them in a way so that we remember them, or check in to see if we're living up to person we think we ought to be.
An example of a character lapse is that I've gotten hooked on the X feed in the last month, and this only happened because I lost sight of a principle. Back in December, that principle was clear. I got a new iPhone, and decided not to auto-transfer my apps. I set it up to be only a "capture device," meaning it had no feeds, no email—just a note pad, a camera, and a voice recorder (on the home screen). It worked well. I had no ill regard for phone. But 10 months later I forgot the principle; September felt like a cultural moment, a COVID-like, cyber-pandemic situation, and so I installed X, thinking it was worth it to be informed. I forgot the consequences, the reason the principle existed. My screen-time went up by 2-3 hours a day, and I bookmarked hundreds of posts, bullshitting myself, assuming that I'd eventually do something with them. Simultaneously, my year's creative output dipped. It took weeks to make the connection, and another few weeks to delete the app.
Maybe it's not worth overthinking this. "Don't scroll" is easy. And isn't your character forged in your childhood? Wasn't I handed down a series of maxims ("be a good person, choose your friends wisely") that shaped me at the tree's trunk? Yes, but also, I think we underestimate the degree that we "sin"—we take dozens of little actions that, are not so terrible, but put together, sabotage the wisest version of ourself and drain us of our life force.
Throughout a year, there are probably 1,000 new insights on ways to improve yourself. We forget most of them. Many of them are also bad, wrong, or misconstrued. In some sense, anything important enough keeps emerging until you internalize it. Maybe that's good enough, but I do think conscious effort is required if you want to push forward into the frontier of your character. (And what is a more worthwhile thing? Forget goals and projects, but guiding who you become is a meta-frame that shapes your whole reality. Character is totalizing.)
Chance are, there is some little epiphany from last week that you forgot, but if you actually remembered it, clung to it, prayed to it every day, and lived by it, it could radically change your life. The problem is it all just slips away. Insight is fragile, and needs a superstructure or nourishment to survive.
A potential method:
Conceive: Pay attention to your inner thoughts throughout the day, and log them. You don't have to write in perfect prose, but aim to capture a full record of anything that has potential. My goal is something like 2,750 words per day in logs (a million words a year)
Review: At the end of the day, read/skim through your day's logs, and save anything that seems like it's worth remembering/changing/acting on.
Loop: Before the week starts, get clarity on what matters; then review your favorite logs, and consider the idea(s) you want to loop/internalize. I don't know the active amount of ideas to be in a loop, but I guess it's something between one and ten.
Prayer: Multiple times a day, pray to those ideas. The word prayer doesn't need to have a religious connotation (though, often it can and should). I see the word as the opposite of "meditation," which is about the cessation of thought. Prayer is about the singularity of thought, on whatever idea is most prescient to you. It covers everything from looping a phrase with eyes closed, to conjuring somatic feelings, to reflecting on a single idea as you walk.
Write: From your set of ideas in the loop, a few might expand more than others. Write about those. That's the material worth unpacking in an essay. An essay after all, should be a vehicle to explore the virtues, questions, and ideas that are most relevant to your life.
Coin: An essay isn't done until you coin it into a maxim that represents the key insight you've arrived at once you're done writing. This compresses the process into a new "prayer" you can loop.
Remember: I've written a million words, and forgotten most of it. The insights gradually fade. The goal of an essay is not just to write a thing, but to extend an insight through time, and part of that is augmented when you review your coined maxims. You can't act from a million words, but you probably can from 100 maxims. This is where something like a hypnomnemata comes in, a book of wisdom that you frequently skim through. Perhaps the maxims are categorized, or summoned based on your journal entries via AI, or who knows. An archive of writing is great to have, but it's perhaps only useful to the degree that you return to your ideas and let them continuously shape you.
I don't have any method like this currently. Now that I've written this, it feels too complex and daunting. But without any system, character is bottlenecked by memory, so there has to be some middle ground. If I were to collapse that whole thing into a single maxim it's: "An essay isn't done until you coin it into a maxim you pin to your wall." Post-its will suffice.
I'm 9 drafts into my Didion essay, and don't feel close to knowing what I really think. Should an essay be moralistic? I think that's the question I'm circling.
Early in its history, the essay was almost indistinguishable from moral writings; they followed a format of A (this thing happened) and B (therefore, this is how we should live). I do think there's value in each person deriving their own morals, yet also, feel there is probably value in cultures having shared standards, expectations, and a sense of what is right or good.
When I disagree with Didion, there a few ways to interpret it:
(A) she is bad/wrong/evil, meaning, in the pursuit of her aesthetic and her career, she has adopted ideas that I think are bad for my character, her character, or anyone's character, whether she's aware of it or not;
(B) she's mostly a good person, but morality is complex, and she has certain facets that diverge from my model of what "good" is (maybe from her model, I'm bad/wrong/evil);
(C) she's good and aligned with my beliefs, but I've misread her... the challenge is she rarely explicitly states what she stands for, and in fact, often says "nothing makes sense and everything is meaningless"
Point (C) brings up something I try to remember as I write: even if I use clever subtext to conceal something, by the end of the essay, you should know where I stand on moral/catalytic matters. Maybe that's just my preference, my attempt to articulate what I mean in precise language, so both you and I know what I think. Didion might just prefer to let readers come to our own conclusions.
But what are her conclusions? Should we celebrate incoherence? Is celebrating incoherence is immoral? I can dabble in chaos, but is all chaos and no order bad? Would she disagree with the notion that she "celebrates" chaos? Maybe she hates chaos, or she acknowledges it without wanting or promoting it, and my sense that she "celebrates" is just my interpretation because she never actually tells me what she believes. It all can get quite circular and ambiguous, which is fine, and perhaps part of the fun.
In contrast, the Essay Architecture system is straightforward in terms of objective patterns we all can see; interpreting virtue feels like dealing with gas. I don't necessarily want to make a periodic table of morals... or should I? (Actually, I have done that, but that's for me personally, and it's hard to know how any of it translates onto others).
I disagree with George’s take on AI here, but paradoxically support everything he has to say about writing, especially this:
“Struggling through reading or writing a long-form piece does magical, essential things to one’s brain, i.e. to one’s essential personhood.”
Can deep effort and AI be reconciled? I think so. You can use AI in your process without having it write a word of your prose. If you use it with intent, it can make the process longer, harder, and filled with weird or delightful frictions. Of course, this is not how most people use it. For most people it’s a shortcut machine. But if you have enough taste to know that today’s AI prose is flat, and enough experience to know the value of toiling through multiple drafts (for the piece, and for you), then it’s possible to use these tools without having them corrupt you.
He says AI is, “dangerous, because, as we are now finding out, it affects the brain and lessens our capabilities.” He may be referring to the 2025 paper called “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” Most people don’t read the actual paper. They ignore the small subgroup of students who drafted unaided but then used GPT to guide revisions, which showed increased “neural connectivity and engagement, with memory recall and activation patterns.”
Basically, I only advocate for AI as an editor. I can imagine George’s response, insisting that the whole point of feedback is to understand how another human responds to my ideas. He wouldn’t be entirely wrong. Whenever I have a draft, I try to get impressions from a wide range of people: my wife, different editors who obsess over craft, general readers who never think about words, and people close to the subject who know things I don’t. AI is just another perspective. Out of the box, it gives you mass opinions (which can sometimes be useful, even as foils). But you can also use it to rotate ideas in personal and idiosyncratic ways. AI is best when it prompts you. I almost never use it to help me write a sentence, but am often using it to try to break my frame of thinking so I can reconceptualize my idea in a more specific, more compelling way. I’m currently 9 drafts into an essay; AI just shattered my thesis, and now I’m going to start from scratch on v10, based on an approach suggested by a friend. I missed my deadline because of AI. Very inefficient, but it will make the essay better.
I also agree with George on the value of printing out your draft and editing with a pen. That’s one mode of many modes. Even handwriting is a technological augmentation, and sometimes I’ll regress to composing with voice. I find value in moving up and down the stack, from voice, to handwriting, to typewriting, to word processing on a computer, to using AI in highly experimental ways. To quote a quote from his piece: “No worthy problem is ever solved in the plane of its original conception.”
It’s ultimately about balancing the new with the old. While it’s extremely handy to have all my past notes and Kindle highlights accessible through a mechanical subconscious—where they emerge in the margins as I write new sentences—it would be foolish to have that be a total replacement for my actual subconscious (the ability to forge new images and associations in the moment). It’s not either/or, but both. Any system that destroys presence isn’t a system worth keeping.
Everything I’m suggesting probably only makes sense if you’ve put in several thousand hours into analog creative processes. When it comes to schools, I basically think AI should be banned. Or at least, high schools and colleges should have structured assignments and exercises that require students to shift between analog and technological modes.
There are many reasons to hate AI, whether it’s the slop, the scam literary prizes, the fatigue from endless debate, or the rate of progress. The existential context fuels a moral crusade against writers who like AI. But I wonder, for all the writers who hate AI, how many of them secretly loathe the editing process too? What is a more reliable indicator of virtue, the open-ended tool you use, or your willingness to struggle through an idea? If you care foremost about the quality of your work and the range of your process, then does it make sense to shut off your curiosity to a powerful, flexible tool you’ve barely experimented with? If you’re thoughtful about it and honest about its limitations, it could take you further.
Tech nerds tend to rush into technology and use it in boneheaded ways, while romantics tend to resist change and cling to older paradigms. My advice is to do both. Pay $20/m for a Claude Pro plan and give Opus 5.5 an earnest try for editing. Also go on Etsy and buy a used typewriter for $300. We’re about to see a fork between accelerationist and retro movements. It serves you to understand the value in each.
A tool isn’t inherently virtuous or not; how you use it is an expression of your virtues.