This is Part Two of a three-part essay series about using AI to supercharge your writing process without sacrificing—well, anything, ideally! Read Part One here:
Also, I’m publishing this essay almost a week late. And there’s no excuse for that. Well, maybe one excuse actually—we just had our third baby last week. Woo hoo!
This is your brain on nail guns
Oh, AI in writing. So fraught. So widespread. A brave new world in a Pandora’s box—but is it bringing the fire of Prometheus, or is it a nude pop star on a wrecking ball, destroying everything about art that is good and holy?1
Folks, for once in my life, I genuinely believe I have some answers. I’m not here to wring my hands or virtue signal or bolster the value of my (nonexistent) shares in OpenAI. Instead, I want to help you navigate this new tool, so you can harness its breathtaking power without letting it dull your humanity.
When you write, you’re building a house. Incorporating AI in your process is going to upgrade you from a hammer to a nail gun—which is amazing news. But you also have a new problem, which is making sure you don’t accidentally shoot a nail into your skull.
To that end, I’d like to propose three AI Commandments. I believe that following these will both empower and protect you as AI continues to weave its way into everything we do.
But let’s face it—I’m no Burning Bush. So first, I owe you a little explanation of how I derived these proposed directives.
The Real Person Test
How lawyers make sense of the brand new
AI may be amazing, but it’s definitely not unprecedented—at least not in terms of disruptive power. The automobile. The internet. The airplane. Clap-on, clap-off lights. Each of these has shattered a pre-existing paradigm.
Every time this happens, lawyers, judges, and lawmakers grapple with one urgent question: how the heck should we regulate this stuff? Without rules, things go Wild West in a hurry, and that means a lot of gunfights and dead cowpokes. But how do you govern something you don’t yet fully understand?
Your best shot in the early game is: analogize to something you do understand.
Bitcoin hits the scene? In some ways, it’s like a security—let’s try applying those rules. Social media becomes the first stop for information dissemination? Maybe we hold it to the speech standards of a newspaper, or a town square. Uber overhauls transportation? Well, what lessons can we learn from existing taxi regulation?
These solutions are never perfect, but they at least give you somewhere to start as things get sorted. And as a lawyer by training myself, this is my first move when approaching novel problems.
And so in thinking about how to use AI, I’ve come up with an analogy that I think is useful as an initial rule of thumb.
Just think about it like a person
When you’re trying to figure out your philosophy for using AI, I want to suggest an elegant little standard called the Real Person Test. It simply asks:
“Would it be right to ask a real person to do this for me?”
That’s pretty much it! Oh—I should say that our hypothetical Real Person is happy and willing to do whatever we’d ask, whether because we’re paying them, or they’re bored, or you saved their Weimaraner from a house fire. So there's no social friction or interpersonal ledger to worry about.
Now let’s take this test for a spin. Want a real person to proofread for you? Totally fine, that’s what copy editors are for. Want to have a deep conversation with them about your ideas before drafting? Awesome, that’s a writer’s bread and butter. What if I want my person to draft my term paper, and then I just rewrite the sentences myself? No way, José. But could I ask them to point out bloated sentences and suggest punch-ups? Definitely—that was an unskippable step when I wrote professionally.
Now, let’s pop the word “AI” into those examples. I personally see no reason why any of those answers ought to change. Sure, it’s probably more meaningful and fun to use a human. And talented editors will do a better job. But if one isn’t available, I don’t see any ethical reason2 why you couldn’t let an LLM take that role instead. Arguments to the contrary seem to fall afoul of the ad hominem—ad robotinem?—fallacy, in fact. Does the truth or quality of information depend on the virtue of its source or not?
So that’s why I spent so much time in my first essay of this series examining how professional writers write and what their support systems look like: because it gives a blueprint for applying the Real Person Test. If you want more insight into that world, check out that essay, linked here.
With the Real Person Test as a backdrop, I have derived three AI Commandments. (I’d call them “suggestions” but that wouldn’t get clicks.) These are especially for creative writers. But not just creative writers: I think all kinds of writers can get value from these. And actually, not just writers: I believe anyone using AI can take something valuable from them.
The AI Commandments
The First AI Commandment: Thou Shalt Always Be the Architect
A writer writes their own ideas, not someone else’s
Right now, my wife and I are working with an architect to design our dream home for our growing3 family. Before any plans were made, we filled out an extensive questionnaire about what we wanted out of a home. Then the architect went in and designed us a home that met our parameters. We were the prompters, and they created the blueprints.
Don’t do this with AI, writers.
Because for all our ideas, we didn’t design that home. We didn’t have to grapple with the layout of our plot, or think through the flow from room to room, or understand how the window placement would affect both interior light and the view from the curb. And that’s why the architect’s name is on the blueprints, not ours.
Here are a few practical applications of this Commandment:
I strongly recommend that you come up with your own ideas. Don’t let AI tell you your next topic to write about, or generate characters for you out of whole cloth, or otherwise hand you your building blocks. Doing so is not only going to limit your vision, but will likely suck the life out of your projects. I find that the amount of passion that I’m able to muster for other people’s ideas is a tiny fraction of the excitement I naturally have for my own.
Preserve your voice—not just in your words, but in how you think. AI will spit things out that, while logical and coherent, tend toward the boring and predictable. Humans are so much better at orthogonal thinking and surprising but meaningful connections.
Remember who’s steering the ship.4 If AI gives you a suggestion you don’t love, either (1) push back or (2) reject it. You’re a human being for crying out loud. It’s a glorified spreadsheet, albeit a really terrifyingly impressive one. Look at me: you are the captain now! And you always were. Choose your own destination and set your own course. Keep your hand on the helm and keep your eyes on the stars. And leave the mainsail-hoisting and poop-deck-swabbing to your robot crew.
The Second AI Commandment: Thou shalt not use AI to draft for thee.
A writer uses their own words, not someone else’s
Wait a minute, you say—isn’t that the whole value proposition of AI? That it can generate incredibly smooth, Turing-test-passing prose at the click of a button?
Yep, that’s the value you’re sold. But no! Don’t do it!
I’ll boil the many reasons why into two: one pragmatic, and one philosophical.
Reason 1: It does a bad job (high floor, low ceiling)
As a generative tool, AI has a surprisingly high floor, but a pitifully low ceiling. What I mean by this is that AI is incredible for creating coherent, mistake-free prose. At least at the level of an average college graduate, and often much better.
But to be known as a writer, you can’t just write as well as a decent college graduate. You need to be orders of magnitude better than that—ten, a hundred, a thousand times better. Most senior theses aren’t getting published in the New Yorker, after all. You need more voice, more originality, more intrigue, more taste. In a word: more humanity.
And remember, great art’s job is to react to and break patterns. AI is just pattern. It’s the difference between a horoscope that’s vague enough to be resonant and a letter from your mother that tells you things about yourself that you didn’t even know. Which do you want to be? What does the world need more of?
Reason 2: Isn’t the whole point to be…a writer?
Pragmatics aside, why would you want to outsource your writing process to a tool?
Oh wait, I know: because writing is horrendously difficult. Right.
You know what else is really hard? Bearing and raising children. And maybe we could outsource that too! You could arrange for a sperm donor to get connected with a surrogate, and then when the child comes, you can place them for adoption by a set of eager parents. That way, you’ve raised a child without having to inconvenience yourself at all!
Think about it: No late nights, no tantrums, and you still get baby pictures to show off. I see no downsides.
Do you?
Look, it’s a free country. Outsource whatever you please. But know that you can’t hire a robot to work out for you and expect a healthy and strong body. And if you want to be a writer, you have to load the oppressive weight of a blank page onto the bar and squat it until you want to die.5 Because that’s the misery, and it’s the reward.
The Third AI Commandment: Thou Shalt Increase thy Cognitive Friction—Not Decrease It
Great writers seek refinement from others, not shortcuts
This is my favorite commandment of the bunch. When you have a new tool like AI, it’s tempting to want to use it to make your life easier. Because the capability is certainly there. There are a lot of difficult tasks that you can offload if you choose.
But this commandment is about leaning into the difficulty. It’s about using AI to test you. To critique you. To question, challenge, and push you. It can add friction, rather than smooth it out. Real readers worth their salt make your writing better by making your life harder. AI should do that as well.
Think of your AI like a personal trainer, pushing you to squeeze out a few more reps when you’re tired. Or a coach who can watch your game footage and tell you where you need more intensity, more focus, or a different strategy. Maybe it’s your piano teacher, listening to your recital and circling spots that still need practice in the score.
With people like those in your corner, success becomes much more attainable. But none of them has made your life “easier” in the usual sense. They create friction, because friction makes you stronger. Lean into it—that’s where the growth is. Stick your whole face right into the refiner’s fire.
So those are my three AI Commandments. I believe that any writer who follows these commandments is going to find their writing process supercharged, at zero cost to their uniqueness, their voice, and their humanity. You’re going to build better houses, faster, without turning yourself into Pinhead with your nail gun.
Nuance and Caveats
Now it’s time for everyone’s favorite game show, Nuances and Caveats!
“It’s not gonna kill me to draft this soulless email with AI.” You’re probably right. And I’ve had ChatGPT draft low-stakes stuff for me before. I think there are all kinds of writing contexts where voice, humanity, and even authorship don’t much matter, so use AI all you want. I saw a Substack note a while back (can’t find it now) that delineated between “writing as a means to an end” and “writing as the end itself,” a distinction I liked. This essay is primarily for people who are writing as the end itself. But be alert—having a unique voice may be more to your benefit than you realize, even in that soulless corporate email.
“I want to deliberately use AI in an experimental or creative way.” Cool! I’ve come out strongly against letting AI generate ideas or draft for you, but I think really what I’m addressing is doing so behind the scenes and uncredited. I’ve read experimental pieces that use AI more extensively and creatively, and I think these kinds of experiments are fascinating and very fun. That said, as a reader, I want to know what I’m reading. Transparency is important. If AI deserves a co-author credit—credit it.
“A lot of stuff in the real world is already ghost written.” This is true. The fact is that while ghostwriters are obviously immoral in an educational context, they are used quite extensively in real-world contexts like celebrity memoirs, legal briefs (drafted by associates, claimed by partners), or scholarship (research assistants drafting for professors). Truth be told, I’ve always found this a little sketchy. But I suppose if you were going to use a human ghostwriter anyway, you could use AI instead. But remember what I said before—outsource this stuff at your own peril.
“These ‘commandments’ feel a little draconian, bro.” What, just because I framed them in the same language as divine mandates? Some might find my recommendations here a little strict, and that’s fair. I think applying these commandments is sometimes more an art than a science, and certainly more a science than a dogma. But if you’re wanting to dip your toe in, start here and get a feel. Better that you start cautious and slowly relax than that you immediately start thinking of yourself as just a “prompter.”
“Wait—if I let AI proofread for me, I’m making my life easier, right? What about the Third Commandment?” I do think you should let AI handle (or help with) more menial labor or rote tasks, like proofreading.6 While this does make your life “easier,” what you’re really doing is freeing up time—time you can reinvest in your writing process. So you’re making your life easier just to make it even harder. Boom!
“Writing in a small cabin by a lake, unaided and untouched—whether by society or technology—is more meaningful to me than anything you’ve experienced in your puny, sad little life.” Hey now, you don’t have to—wait a second, is that you, Henry? Ah ha! Well if it isn’t Hank Thoreau himself. Look: I think you should absolutely use AI as part of your writing process. But if you are a writing ascetic, then more power to you. People seem to find immense value doing all kinds of mind-boggling things, like running ultra-marathons or doing juice cleanses or having three children. My pitch is: choose it consciously for your own reasons, rather than out of fear of something new that you don’t yet understand.
Prometheus and s’mores
I said before that AI wasn’t unprecedented. But either way, it is wild. And I, for one, am thrilled about it. I feel like we’ve only begun to scratch the surface of how transformational LLMs are going to be for us in the long run.
But I also get that new tools are scary. Some of them are powerful enough to put nails into your brain. But when Prometheus brings you fire, the right response isn’t to pour dirt on it and go back to eating berries and twigs. Better to experiment—deliberately and carefully, yes, but also with enthusiasm. Get to know its properties and understand how it can be useful and dangerous to you. Figure out what’s most important to you and make sure that you’re protecting that first. That way, even if you get the occasional singe, at least you won’t burn down your village in your excitement to make tastier meat.
As a creative writer, what’s important to me are my voice, my ideas, my words, and my reputation. I personally keep the fire a good distance from those, by using the Real Person Test and the three AI Commandments I’ve proposed here. But with those guardrails, I’m having the time of my life burning stuff. And I’m learning a lot, too.
So now we have some ground rules. In my next and final(?) essay of this series, I’m going to walk you step-by-step through my writing process and show you exactly how I prompt and engage with AI as I do so. If you’d like to join me, grab a stick and a marshmallow and let’s see what we can cook with this thing.7
I dare ChatGPT to mix that many metaphors in one sentence.
From an authorship and credit standpoint, that is. I understand that there are lots of ethical arguments against AI broadly and the companies that provide it—environmental, copyright, etc. But I don’t know enough about those arguments to comment on them, and anyway it’s outside my scope here. I’m not trying to tell vegans they should eat meat—I’m trying to help meat-eaters understand how to cook it.
New baby! Woo hoo!!
I know I’m supposed to be on architecture metaphors, but right now I’m feeling nautical.
Is “last-second metaphor switcheroos” a writing style? Because I think I’m nailing it.
Real copy editing nerds will argue that proofreading isn’t rote, but is itself an art. They’re absolutely right. (I see you, baby.) But for most of us, it’s an art that is tangential to our goals as writers at best, and a distracting slog at worst.
Wait—I should be ending with a nail gun callback! Dang it. Curse you, metaphor switcheroos!
This post has not gotten NEARLY the amount of love it deserves.
This one's worth stealing.
I don't agree with your conclusions (I think gAI is more harmful than helpful) or your attitude (opposition to these services are not largely based on fear-- as in being scared of what will happen or worried about it from ignorance of how it works-- but on real considerations of problems with it; to say nothing of my distrust of proponents of the technology), but I think anyone interested in using gAI should definitely read this article for some things they should definitely be considering before trying it.
I recently saw a note from an author who gave anecdotal evidence of a tendency toward changes in how their thinking worked (with diminished capacity) after using these tools. It meshes with what you mentioned about making things easier instead of more efficient or strengthening. I don't think you put enough weight on the dangers.
That said, I probably will restack this after figuring out how to introduce it.
I enjoyed reading it, up until I got to the nuances and caveats, which I thought worded the impact of your cautions in the main article.