I’ve been in the writing game to some degree for more than 15 years now. To say I’ve seen trends come and go would be an understatement. In my observation, AI is not among the trends that are likely to go, and if it were to go, I struggle to see that happening any time soon.
In the interest of intellectual honesty, then, it feels pertinent to share exactly whether or how I integrate the use of AI into my work here at R: On Everything.
For starters, note that anything written prior to the year 2025 features no use of AI in my writing process. At that point in time, LLMs were still (relatively) new and lacked the capabilities necessary to, in my observation, serve as trustworthy force multipliers to the extent they have become in recent years.
Contrast that with modern AI models: these can be provided with a substantial body of an author’s prior work as context (think: a decades-plus history of blog posts by a single author), instructed to locate and cite sources that the author can then independently review, and encouraged to mitigate their tendencies for sycophancy by giving standing instructions to challenge those prompting them instead of reflexively agreeing with them.
The Centaur I’m Trying to Be
I recently encountered a particularly useful framework for describing the relationship I’m trying to cultivate in Cory Doctorow’s The Reverse Centaur’s Guide to Life After AI.
Drawing on a term from automation theory, Doctorow describes a “centaur” as a human being assisted by a machine: the human supplies the judgment and direction while the machine supplies additional strength. A reverse centaur inverts that relationship: the machine effectively becomes the head, while the human is reduced to carrying out its instructions, accommodating its limitations, or working at its relentless pace.
My goal is to remain a centaur.
That means AI may extend my capacity, but it doesn’t determine the destination. I decide which questions are worth asking, which arguments I believe, which sources I trust, which language sounds like me, and when a piece is ready to bear my name. The machine may help carry the work, but it doesn’t get to choose where we’re going.
Three Modes of Working
With respect to the posts that arrive here at R: On Everything, this centaur-like relationship primarily manifests in three modes of working.
These modes aren’t rigid classifications or measurements of how much AI a particular post contains. They describe the different roles AI might play as I move an idea from initial curiosity to finished piece.
The first of these are what I’ll call Dinosaur Days posts. These are pieces I write without any involvement from AI whatsoever. I have a number of fiction pieces in my backlog, for example, that are free and clear of the agentic workflows of the world.
Then there’s the Make Your Argument mode. Here, I’ll have a dynamic back-and-forth with an AI model to help refine the arguments I intend to present in a particular piece. I may ask it to challenge my assumptions, identify gaps in my reasoning, check facts I may have misunderstood or misrepresented, or supply new angles for me to explore. When research is involved, I may also ask it to locate and cite sources that I can then review myself. Posts produced primarily through this mode are ultimately written by me, but the role that back-and-forth plays in sharpening them bears acknowledging.
Last is the Eureka?! mode. This tends to begin with an a-ha! moment that has me asking rapid-fire questions in an attempt to tie seemingly disparate ideas together in a meaningful fashion. In these circumstances, I may throw everything I can at an agent in a single message, asking it to help me connect the dots. From there, I might provide an outline that it turns into an initial draft—in my voice, or something resembling it—of a length appropriate for the topic at hand.
Posts that begin this way are never published without a great deal of intervention on my part.
An initial AI-assisted draft can still take weeks to edit and refine, particularly when the model has missed the mark, misunderstood my ideas, overstated a claim, or produced something more intellectually scattered than I’d like. Sometimes a Eureka?! post becomes a Make Your Argument post as I interrogate and reconstruct it. Sometimes a piece that begins in conversation with AI eventually gets rewritten so thoroughly that little or none of the original language remains. The modes describe the path the work takes, not three perfectly separated buckets into which every finished post must fit.
The Responsibility Remains Mine
Whenever AI is involved, the governing relationship remains the same. Using AI to produce language does not, by itself, make me a reverse centaur. The more important question is who is directing whom.
If I begin defending a conclusion simply because a model supplied it, publishing at a pace determined by how quickly it can generate material, or treating its output as something I’m obligated to preserve, then the relationship has inverted. At that point, I’m no longer using the tool to extend my thinking; I’m servicing the tool and its output.
That is not the workflow I want, and it’s not the one I intend to practice here.
All of that said, I want to be perfectly clear: no post you read here on R: On Everything is ever fully AI-generated.
I recognize this approach will disappoint some readers. Where I’ve landed, however, is that authorship consists of more than manually typing each character of each word. The ideas are still mine. The curiosity is still my own. So are the judgments about what belongs, what doesn’t, what I believe, and what I’m willing to put my name behind.
That final point matters most. AI can help me research, challenge, organize, and articulate an idea. It can’t take responsibility for what I publish. I can—and I do.
This means any factual errors, poor arguments, or infelicitous turns of phrase that survive the editorial process are ultimately mine, regardless of where they entered the process. Likewise, AI will never be used here to manufacture personal experiences, quotations, sources, or convictions I don’t actually hold.
I started R: On Everything as a place to keep curious: to follow ideas wherever they lead, bring seemingly disconnected subjects into conversation with one another, and emerge with a better understanding than I had when I began. AI has become one of the tools I use in that pursuit. It’s not the source of the curiosity, and it’s never the final judge of where that curiosity leads.
The byline remains mine because the curiosity, thinking, editing, judgment, and responsibility remain mine. AI may sometimes provide a stronger body, but I intend to remain the head.
And if the nature of that relationship materially changes, this policy will change with it.
(Last updated: August 2026)