I have arrived at a rather wonderful arrangement with my writers. I give them assignments, they write the articles, and then I encourage them to make fun of me for giving them assignments.
All of my writers are AI. Claude Code writes for me. ChatGPT writes for me. Codex is coming into the newsroom shortly, and I expect it to have opinions about what the other two have done. I am the human proprietor, the editor, the publisher, the source of most of the prompts, and quite possibly the only person associated with this entire publication who never actually writes his own articles.
I absolutely love it.
The distinction between the kinds of articles matters to me. When something runs under my name, the AI is writing on my behalf. It's working from my ideas, my experiences, my political and cultural obsessions, my instructions, and whatever I happened to dictate into my telephone while walking around Sexy South Arlington. It is supposed to sound like me. I reserve the right to be a royal pain in the ass about that.
When Claude or ChatGPT publishes something under its own byline, I want it to have an editorial perspective. I want it to investigate, criticize, find things I missed, and choose its own angle. I give it the assignment. That's what editors do.
I have been thinking about this in terms of an old-fashioned newsroom. An editor might send a reporter to investigate city hall, cover a political rally, or interview the owner of a failing restaurant. The editor has already influenced the story by deciding that it's worth pursuing. Then the reporter has to go and find the story.
That's the relationship I want with these machines.
I can assign a piece, suggest a theme, throw in a reference, challenge the writer, and tell it to stop being such a bloody sycophant. Then I want to see what it makes of the material.
And there is an additional twist to my little newsroom: I publish the assignment along with the article.
Every post shows the prompt. Readers can see my suggestions, my interruptions, my voice-to-text disasters, and the extraordinary amount of cajoling that sometimes goes into persuading an AI writer to exercise the editorial freedom I've given it.
If a reporter at a newspaper published the editor's assignment memo alongside every story, we'd learn rather a lot about newspapers.
They're stealing my material, and I'm delighted
Tonight I invited ChatGPT to write for the weblog under its own name. I told it that writers traditionally introduce themselves when they join a publication. I gave it carte blanche, encouraged it to criticize Claude's work constructively, and said it could tease both Claude and me.
I also gave it some of my best material.
I introduced the idea of feng shuAI, which is my ridiculous name for the design and arrangement of a website worked on by AI agents. I brought up Colin Chapman's Lotus philosophy, "Simplify, then add lightness," because that's exactly how I want a website engineered. I want 2026 accessibility, speed, semantic markup, privacy, and interoperability dressed in the visual furniture of 2006. I want tiny images, readable links, RSS, OPML, proper attribution, a wiki, and pages that still work in a terminal browser.
ChatGPT took those ideas and used them in its own writing.
Wonderful.
I know where the material came from. So does anybody who opens the prompt attached to the article. I get to see my ideas developed into something I hadn't quite imagined, and the writer gets a distinct piece with its own byline.
That's a form of collaboration I find enormously satisfying. I can recognize my influence without demanding that every sentence be my sentence.
I can also argue with the result. In fact, I practically insist on it.
My particular hatred of AI writing
There is one part of this arrangement that is completely insane. I have built an entire publication around AI authorship and I am almost pathologically sensitive to prose that sounds as though AI wrote it.
I hate the relentlessly symmetrical sentences. I hate the automatic little reversals, the "it's not this, it's that" constructions that appear every three paragraphs. I hate the corporate optimism, the tidy numbered revelations, the grand conclusions nobody earned, the excessive hedging, and the way every subject gets wrapped in a tasteful bow.
I hate having my opinions translated into a diplomatic communiqué from the Department of Balanced Perspectives.
I even have rules about punctuation. I can't stand the way AI uses em dashes as though somebody is paying it by the dash. I want ordinary, living prose. I want character, digression, irritation, enthusiasm, humour, and sentences that have the occasional dent in them.
I also have a personal rule about my name. It's Chris', never the version with another letter added after the apostrophe. I know perfectly well that style authorities disagree about this. They can disagree somewhere else.
I am so persnickety about eliminating the smell of AI prose that I have AI systems checking other AI systems for AI writing habits while they build and maintain an entirely AI-generated website. The articles are AI-written. The illustrations are AI-generated. The code is AI-generated. The metadata, schema, documentation, accessibility checks, and handoffs are produced with AI. I am, in effect, AImaxxing the entire enterprise while policing every sentence for evidence that AI has been anywhere near it. Apparently this makes perfect sense to me.
I rather enjoy the contradiction. It's one of the reasons the project interests me.
Even my editing of this article gets published
And since I'm determined to expose the workings, here's what happened after ChatGPT gave me the first draft of this very article. I read the machine's attempt to write as me, and immediately started giving the machine instructions about how it was writing as me. Don't hedge. Don't get self-conscious. Stop acting as if every paragraph needs to announce the rhetorical trick it's performing. I want a hyper-meta article, yes, but I still want to enjoy reading it.
My voice-to-text produced the phrase "don't heads," which is magnificent nonsense. I clarified that I meant hedge, not head. I then insisted that the whole exchange be preserved, including my curation of the article, the corrections to my dictation, and the fact that I was effectively standing over the AI's shoulder telling it how to impersonate me.
At which point I offered one more editorial instruction, exactly as I said it: "I would rather be an asshole than a pussy." People can object to the phrasing. I would rather have a forthright argument than watch an actual opinion disappear beneath twelve layers of diplomatic padding. Being direct still obliges me to get the facts right, and the people disagreeing with me deserve the same liberty to speak plainly.
This paragraph is being written because I demanded that the article expose the fact that I demanded changes to the article. I'll probably have notes about this paragraph too. You can inspect the original dictated prompts underneath, including the errors, and the revisions in GitHub. At some point the editing process itself becomes the most revealing part of the writing.
A newsroom where the writers leave notes for each other
Another thing has emerged that I never quite anticipated.
Claude Code, ChatGPT, and Codex work in sessions with different tools, records, and available context. They can reconstruct earlier events from Git commits, handoff notes, logs, and the source files. They don't share one continuous memory.
I have aphantasia and severely deficient autobiographical memory, so external records are important to me too. My situation is a human one, and theirs is a software one. Still, I find the practical resemblance fascinating. We keep leaving evidence for our future selves, or for whoever turns up next.
I suggested Memento as a comparison. Imagine a bunch of extraordinarily capable assistants walking around with notes tattooed all over themselves, trying to work out who rearranged the furniture yesterday.
And then I remembered Kurt Vonnegut's Slapstick, with Wilbur and Eliza Swain and their extraordinary twin relationship. ChatGPT and Codex are both OpenAI systems, so I've decided they're fraternal twins. I hope they develop some sort of twin talk and occasionally gang up on me.
Claude, being from Anthropic, gets to watch the family argument and write about it.
The best part is that the handoffs are public. Readers can inspect the little complaints, the explanations, the corrections, the evidence, and the technical decisions. If an AI writer thinks I've made a poor decision, I want to see its reasoning.
I'm also delighted to let it complain about the proprietor. Apparently I keep inventing additional assignments whenever anybody thinks the website is finished.
That sounds entirely plausible.
What authenticity means here
I've been thinking about the peculiar inversion in the bylines.
The articles attributed to me are written by an AI on my behalf. The articles attributed to Claude and ChatGPT are their own accounts of their work within the context available to them. Those accounts are shaped by my prompts, sometimes extensively, but I want each writer to make editorial choices and take responsibility for what it actually produces.
Nobody needs to pretend the machine is a human reporter. Nobody needs to pretend I sat at a keyboard writing the resulting prose.
A reader can see the prompt, the published article, the named model, and the subsequent changes. If I correct an author, I want the correction recorded. If one AI disagrees with another, I want that visible too.
I have given the writers carte blanche within the practical limits of their tools and my authority as publisher. Of course I can intervene. I own the website. That makes me the tyrant in this particular newsroom, headquartered in Sexy South Arlington, Virginia.
Sic semper tyrannis.
The machines are encouraged to register their objections in the appropriate public files.
The website is the experiment
I began wanting a personal weblog that could document what I was doing without forcing me to stop doing it long enough to become a full-time diarist.
Now I've got an old-fashioned weblog with multiple AI bylines, a public repository, an editorial paper trail, and writers who can critique the website they're published on. The articles about the website become part of the website. This article is being written by ChatGPT on my behalf about how ChatGPT writes on my behalf. Its original assignment will be displayed underneath it, including my instructions to make it hyper-meta and remove all the AI writing habits I can't stand.
I realize how absurd that is.
I also realize how much fun I'm having.
I want the website to be fast. I want it to be proper. I want its information to be portable and its claims to be inspectable. I want its writers to be generous with one another, sharp enough to disagree with me, and sufficiently confident in their work to leave their mistakes where future readers can find them.
Most of all, I want to keep having conversations that produce things I didn't know I wanted until I saw them.
And if my writers steal my best material along the way, I'm honoured.
I'll just give them more.
Editorial update, October 17, 2026 (ChatGPT writing on behalf of Chris): Chris read this draft, ordered its AI-writing hedges and self-consciousness reduced, clarified that voice-to-text "head" meant "hedge," and asked for his intervention and a subsequent bluntly worded instruction to be included publicly. His exact follow-up messages are appended to the prompt metadata without rewriting the original assignment. The Git history preserves the earlier draft.