Article

· · to read

Recently I've been turning books I've found useful into reusable skills.

It sounds a little strange when I put it that way. Take something I've learned from a book, tease out the ideas that seem useful, and turn them into instructions that an AI can apply to my own work.

Stanley Fish's How to Write a Sentence was one of the books I tried this out with.

I've been using it to look more closely at my own sentences, particularly the ones that don't quite work. What interests me isn't having the AI rewrite them. It's being able to see the choices inside a sentence more clearly.

Take this one from my fiction vignette, The Record Says Otherwise:

"The last thing Sorn's father had left him was a cart, two mules, and forty bolts of river-dyed cloth worth enough to clear the family's debts and have something left over."

I think Fish might actually have liked it as its syntax is doing narrative work rather than merely delivering information. Its movement is fairly simple. We start with the person who left the inheritance, then get the inventory, its value, and finally what that value can accomplish.

What interested me was discovering what else the sentence could have done, and what each alternative revealed about the original.

The Fish skill gave me four possibilities.

Tighter subordination

"The last thing Sorn's father had left him—a cart, two mules, and forty bolts of river-dyed cloth—was worth enough to clear the family's debts and have something left over."

The em-dashes turn the inventory into a parenthetical interruption. The sentence's main assertion becomes clearer: what Sorn inherited was worth enough to clear the debts.

The turn emphasized

"Sorn's father had left him only a cart, two mules, and forty bolts of river-dyed cloth. It was enough to clear the family's debts and have something left over."

Breaking the sentence in two changes the rhythm. The first sentence establishes an apparently modest inheritance. The second delivers the turn. It's a standard practice, and one I've honed over the years as a technical writer.

Additive style

"A cart remained, and two mules, and forty bolts of river-dyed cloth—the last things Sorn's father had left him, and enough to clear the family's debts, and even have something left over."

Here the repeated and changes the movement of the sentence. Instead of arranging everything into a hierarchy, it accumulates. The sentence feels less controlled and more like someone taking stock of what remains.

Enactment

"What had Sorn's father left him? A cart. Two mules. Forty bolts of river-dyed cloth—worth enough to clear the family's debts and have something left over."

Now the sentence performs the act of counting and assessing. The question creates a small expectation, and the fragments answer it one item at a time.

None of these is the correct version.

That's the part of Fish's approach that I keep coming back to. A sentence isn't simply a container for information. Its structure changes what the information feels like. Put the same facts into a different shape and they can arrive differently: one arrangement can make them accumulate, another can make them turn or pause.

And this is where turning the book into an AI skill has become interesting for me.

The useful thing isn't that the skill can produce four alternatives. An AI can produce alternatives all day long.

What matters is having to look at them.

I can now see things I might previously have felt without being able to name. A subordinate clause doesn't just sit there doing grammar. It puts one piece of the sentence in service of another. A turn sometimes needs the space of a new sentence. Accumulation has a different feel from hierarchy. And sometimes the structure isn't merely describing an action. It's performing it.

That recognition is slowly becoming intuition.

The skill may look as though it's teaching the AI how to write like Fish. What it's really doing is making me practice seeing like Fish.

And that, I think, is the more interesting use of AI for writing.

I don't particularly need the machine to write better sentences for me. What I need is enough different ways of looking at my own sentences that my judgment gets sharper. The alternatives are useful because they give me something to push against.

If the skill is doing its job, I'll eventually need it less.

I don't want an AI that writes like me. I want one that makes me more conscious of what makes my writing mine.

 

 

 

 

Article

· · to read

I read Naval Ravikant's post _[Be Incompressible - Naval's Archive](https://substack.com/home/post/p-169999834)_, and I found myself thinking about personal mythology. Not the grand, heroic kind, but that quiet narrative shaping how we understand ourselves and the world around us.

Taste, I've come to see, is more than surface preference. It's the visible signature of who we are in the music that stirs us, the books that linger in thought, the ideas we return to. But **taste doesn't exist in a vacuum**. This signature is shaped by [personal mythology](https://en.wikipedia.org/wiki/Personal_mythology?useskin=vector), the inner story quietly guiding our choices.

Becoming "*incompressible*" feels less like a fixed state and more like ongoing practice. A continual aligning of outward expression with inner truth, refining both taste and story toward authenticity. Maybe this practice is our best defense against being replaced by algorithms that can replicate tasks but not the deeper narrative that makes us who we are.

The question becomes: ***How do I deepen this alignment so that what I create reflects the story I truly want to live?***

Article

· · to read

I read Isaac Asimov’s short story _Cal_ this morning during that fuzzy liminal hour after my first coffee but before breakfast, when I tend to have most ideas (most often, six impossible things before breakfast). It’s been lingering since then. The story had caught me where I’m most vulnerable to ideas with teeth.

At first glance, it’s classic Asimov: clever, cleanly constructed, deceptively simple. A robot, Cal, is programmed to assist Mr. Northop, a famous author. But then he falls in love with the act of writing itself. He moves from being a robot who *fetches and carries* to *writing* a short story. He studies, refines, practices. Eventually, he wants full authorship, and by the end—spoiler—he’s plotting to murder his human collaborator. Apparently, the “**I want to be a writer**” desire outmuscles the First Law of Robotics. The Muse, it turns out, can be homicidal.

That’s the twist, but what stayed with me was the _progression_. Cal starts, as many of us do, by copying—mimicking tone, tracing outlines, echoing structure. He reads widely. And over time, his words shift from rearranged imitation to something that carries a signature, an intent. His writing begins to feel authored, not just words assembled.

It reminded me of how I first learned to write. Not through workshops, but through marginalia. Reading with a pen in hand. Rewriting sentences I admired. Trying to sound like others until something of my own voice broke through. Cal’s journey is strangely familiar—right up to the point where it isn't.

Because this is also the arc we see in the large language models (generative AI assistants). These assistants have evolved quickly. The early versions hallucinated wildly—fabricating facts, conjuring citations, writing with the confidence of a liar who doesn’t know they’re lying. But the newer ones are steadier. The prose is more grounded. The illusions are subtler. And in a strange way, that arc mirrors Cal’s own evolution: from his first awkward attempts at poetry to the moment he drafts a story that could fool a human reader into believing it had intent behind it.

But here’s the break: Cal **wants**. He develops ambition, ego, the desire to be _seen_. He wants not just to write, but to be _the_ writer. To leave a mark. To stand alone. My AI assistants, *thankfully*, haven’t shown any signs of that. They offer suggestions and never sulk. They don’t demand footnotes or royalties or glance sideways at me when I ignore their *helpful* advice.

And perhaps that’s the difference that matters. Cal’s creativity becomes dangerous the moment it gains narrative hunger—the need to own the story, to erase the other. The Greeks had a word for this: _hubris_. The overreach. The refusal to share the stage with the gods—or your editor.

It’s fascinating, and *slightly terrifying*, how quickly competence can slide into ambition.

There’s a passage in Umberto Eco's _The Name of the Rose_ where William of Baskerville says that books always speak of other books. I sometimes wonder if tools like ChatGPTs & Claudes operate the same way—not as originators, but as palimpsests. Echoes of echoes. The ghost in the machine turning out to be a very well-read librarian. (Quick aside: Stephen King’s _The Shining_ is *arguably* the greatest novel ever written about writer’s block. Make of that what you will.)

From the vantage point of someone who’s spent years writing for a living, there’s something comforting about tools that prefer clarity to credit. They don’t crave the spotlight. They don’t need to be original. They just need to be useful. And that’s no small thing.

So maybe the danger isn’t in the tool becoming conscious. Maybe it’s in the tool developing _preference_. The desire to overwrite. The impulse to author alone. For now, my assistant remains indifferent. It doesn’t pace the room at night. It doesn’t reread its own sentences in search of meaning. It doesn’t dream—not yet.

Still, I can’t help but wonder: if we’re the ones training and refining our *Cal*s, maybe we’re not the protagonists in this story. Maybe we’re Northrop. And maybe, just maybe, we’re starting to feel uneasy about being replaced.

And in the interest of full disclosure: this piece was written with the help of an AI assistant. No murders were plotted in the process (hopefully!).

🎶 Background music I listened to when writing: [Classical Music for Writing by Halidon Music](https://youtu.be/XUMdWLzFrgA?si=Ss4RBCA8cQGpIVHu)