When the Friction Is the Work

For most of my life, I thought writing meant choosing the words.

Not merely having an idea and communicating it successfully, but deciding exactly how one sentence should lead into another. Whether this word was better than that one. Whether a paragraph needed one more sentence or one fewer. Whether some thought sounded too formal, too blunt, too sentimental, too vague.

I cared about these things partly because I was lucky enough to begin reading seriously when I was young, and partly because I had English teachers who made writing feel like something worth taking seriously.

Language wasn't just packaging for thought. The packaging was part of the thought.

That makes my current writing process slightly strange.

These days, I often begin with the ideas and outsource much of the composition.

I explain what I am thinking to an artificial intelligence. I describe the argument, the uncertainty, the examples, the tone I am after, the things I do and do not want the piece to say. Then I get back a draft.

I read it.

And surprisingly often, my reaction is simply:

Yes. That's basically it.

Maybe I change some things. Maybe I push back against a sentence that goes too far or ask for a section to be reworked. But if the words adequately represent the structure that was already in my head, I increasingly feel little compulsion to replace them merely so that I can say I personally typed every sentence.

This is extremely efficient.

It is also, if I think about it for too long, a little unsettling.

If You Outsource the Sentences, What Happens to the Voice?

One traditional idea of becoming a writer is that you slowly develop a voice.

You write thousands and thousands of sentences, many of them bad. Certain rhythms begin recurring. Certain kinds of observations become characteristic of you. You develop habits, preferences, weaknesses, tricks.

Eventually, someone might be able to read a page without seeing the byline and have some chance of knowing who wrote it.

But that process assumes that you are actually writing the sentences.

If another system is producing much of the prose, even under detailed direction, then some portion of the resulting voice necessarily belongs somewhere else.

You can prompt for your preferences. You can provide examples of your previous writing. You can edit the result until it feels right.

But there is still a difference between recognizing a sentence as something you approve of and discovering that sentence yourself.

That difference seems important.

I am less certain what it means.

It might mean that heavy reliance on AI eventually smooths away some of the peculiarities that would otherwise develop into a distinctive style.

The awkward constructions you would eventually turn into strengths disappear before you ever write them. The strange metaphor you would have stumbled into never occurs. You don't spend twenty minutes wrestling with a paragraph and accidentally discover that the real subject of the essay is something slightly different from what you thought.

The machine greases the tracks.

But perhaps some useful things happen because the tracks were not greased.

Writing Is Also a Way of Thinking

This is where the question becomes more serious to me.

I am not particularly worried about satisfying some purity test of authorship.

I don't think pressing every key personally gives writing magical authenticity.

What concerns me more is the possibility that composing prose is itself part of thinking.

Sometimes you don't actually know what you believe until you try to write it.

A vague idea encounters the resistance of language. You try to explain it. The explanation doesn't quite work. You try again. Somewhere in the struggle, you realize that the idea needs to change.

Writing creates friction, and friction can be intellectually productive.

If AI removes too much of that friction, perhaps it can help me publish more ideas while simultaneously allowing some of those ideas to remain less examined than they otherwise would have been.

That would be a strange trade.

I could become more productive as a writer while getting less practice at one of the things writing is supposed to teach me.

On the other hand, the opposite can also happen.

There are ideas I probably would never turn into essays if every one required several uninterrupted hours of composition. There are only so many hours in a life, and the requirement that every worthwhile thought must first survive an elaborate ceremony of manual prose construction seems arbitrary.

AI lowers the activation energy.

An idea that might have remained a paragraph in my notes can become a finished essay. A question I would otherwise abandon can be developed far enough for me to see whether it actually goes somewhere.

That matters too.

Perhaps removing friction sometimes impoverishes thought.

Perhaps removing friction sometimes allows thought to exist at all.

The Strange Question of Caring

There is another quality to writing something entirely yourself that has little to do with technique.

It takes time.

To spend an evening writing an essay from the first sentence to the last is, implicitly, to declare that the subject deserved an evening of your life.

That investment can seep into the writing in ways that are difficult to measure.

You linger over things.

You notice things.

You make connections partly because you have been sitting inside the subject for hours.

There is a kind of osmosis involved.

When composition becomes almost instantaneous, that connection between effort and attention weakens.

But effort is not the same thing as care.

Typing slowly is not morally superior to typing quickly.

Someone can spend six hours producing something empty, while someone else can spend twenty minutes directing a tool toward an idea they have been thinking about for ten years.

So I don't think time spent can settle the question either.

Who Am I Protecting My Voice For?

There is also a faintly ridiculous aspect to worrying about one's precious literary voice.

It sounds like the kind of concern an established writer ought to have.

If I were Joan Didion or James Baldwin and a machine suddenly started sanding off the characteristics people had spent decades recognizing in my prose, perhaps there would be something obviously valuable at stake.

But who am I to become territorial about the sanctity of my sentence structure?

Most of the time, my goals are simpler.

I have an idea.

I would like to get it out of my head.

I would like another person to understand it.

Ideally, I would like them not to become bored halfway through.

If a tool helps accomplish those things, it seems strange to reject it merely because writers historically had to do more of the mechanical work themselves.

And yet I don't entirely trust that argument either.

You do not have to be famous before your voice matters.

Perhaps voice is valuable precisely because it develops long before anyone has a reason to care about it.

Hunter S. Thompson and the Myth of the Untouched Page

Hunter S. Thompson is an interesting figure to imagine in this situation because his prose seems almost maximally inseparable from its author.

The idea of feeding a subject into a corporate language model and asking it to produce something vaguely Gonzo sounds almost sacrilegious.

But Thompson's real working process complicates the picture.

His writing was famously collaborative and chaotic. Editors at Rolling Stone sometimes had to assemble fragments arriving under deadline pressure into coherent pieces. Material from tape recordings was transcribed for Fear and Loathing in Las Vegas. Later in his career, assistants increasingly helped organize and turn his spoken material into publishable writing. Peter Richardson’s account of Thompson’s writing process in Bookforum gives a vivid sense of how collaborative and improvised that process could become.

None of this makes the work seem less like Hunter Thompson.

Which suggests that authorship has never required that the author personally perform every mechanical step between thought and printed page.

Dictation doesn't destroy a voice.

Transcription doesn't destroy a voice.

Editors don't necessarily destroy a voice.

So AI cannot be disqualified merely because another entity helped produce the sentences.

The harder question is how much agency over language can be transferred before something essential actually does change.

And I don't know where that line is.

I suspect Thompson himself might have hated the entire infrastructure surrounding modern generative AI even if he found the underlying shortcut irresistible. I can easily imagine both reactions existing in the same person.

That contradiction feels familiar.

For Now, I Keep Using It

At the moment, I don't have a manifesto.

I have a working rule.

If I read something produced with AI assistance and feel that it misrepresents what I think, I change it.

If I begin noticing that everything I publish sounds generically competent in the same way, I will worry.

If I start finding that I can no longer produce good prose without assistance, I will worry much more.

But I haven't reached that point.

Right now, the dominant effect feels less like having my voice replaced and more like having the distance between an idea and a finished piece drastically shortened.

That may carry a cost.

It probably does.

Every technology that removes effort also removes some of the experiences that used to happen during that effort.

The question is whether those experiences were essential or merely incidental.

I suspect the answer will be different for different kinds of writing.

There will probably always be things I want to write slowly, myself, sentence by sentence, because the act of finding the words is inseparable from what I am trying to understand.

And there will be other times when I already know what I want to say and simply want the machinery to move faster.

Perhaps the real skill for writers in the age of AI will not be deciding whether using it is authentic or inauthentic.

It will be learning to recognize the difference between those two situations.

Knowing when the friction is merely friction.

And knowing when the friction is the work.