Maimonides opens his Eight Chapters, the introduction to his commentary on Pirkei Avot, with a confession. The ideas that follow, he tells the reader, "are not of my own invention." He gathered them from the Sages, from the Midrash and the Talmud, and from the philosophers, ancient and recent. Sometimes he'll quote a passage word for word, and he wants you to know that "it is not done with the intention of glorifying myself by presenting as my own something that was said by others before me."
He puts that disclosure first, before a single chapter, so the reader knows how the book was made before reading any of it.
This is mine.
What I tried
I spend a lot of my time building automation, and at some point I pointed that habit at my own writing. I trained a system on my voice and let it draft finished pieces. The question was simple: if an agent could write the way I write, could I spend my time on everything around the writing?
It didn't work, and the way it failed taught me something. In one week the system gave me four candidate posts. I approved two of them. Before either went out, I had rewritten nearly 60% of the text. Each one sounded like me. None of them said quite what I meant. The drafts were stilted and conventional, and a few of them circled the point without ever landing on it.
A familiar turn of phrase can't tell you whether you believe the argument. It can't tell you whether you'd defend the example if someone pushed back. Those are the decisions that make a piece yours, and I had handed them to a process that could imitate my sentences without making any of my choices.
The thief the Torah worries about
The rabbis had a name for a particular kind of dishonesty: genevat da'at, literally "stealing the mind." The Talmud in Chullin 94a forbids it outright, and Maimonides codifies it in Hilchot De'ot 2:6. A person, he writes, should not say one thing outwardly and think another inwardly. "What he feels in his heart should be the same as the words on his lips." He adds that even a single word of deception is forbidden.
Rabbi Shimshon Raphael Hirsch, the nineteenth century rabbi of Frankfurt, sharpens the definition in his commentary on Leviticus 19:11, the verse "you shall not steal." For Hirsch the verse reaches past cash and property:
"Every action, every word, by which we make other people have a higher opinion of us than we really deserve, or gain more gratitude than we have really earned, is reprehensible as being [genevat da'at]."
When I read that line, I thought about every post that goes out under my name. A reader who likes a piece gives the author a little credit: for the insight, for the phrasing, for having done the thinking. If the machine did the thinking and I took the credit, Hirsch would call that theft, even if nobody ever found out.
Harry Frankfurt gets at the same problem from a different angle in his essay On Bullsh*t. What bothered him most about bullsh*t wasn't lying, because a liar still has to know the truth in order to hide it. The bullsh*tter has something worse: "this indifference to how things really are." A model trained on my voice isn't lying when it writes a paragraph I'd never write. It isn't doing anything with the truth at all. It produces text that fits the shape of what I usually say, and whether that text matches what I actually think is a question it can't ask.
That question has to be asked by me.
What the machine does now
So I changed the division of labor. I still use my agentic framework for my writing. It just works at the two ends of the process instead of the middle.
Before I write, it prepares a brief: an angle to consider, why the topic is timely, possible sections and what each might focus on, specifics I can trace back to verifiable sources, and questions only I can answer. I read it and throw out whatever doesn't belong. Sometimes the brief tells me I don't have enough to write the piece yet, and that's a useful answer too.
Then I write every sentence myself.
After I write my posts, a scoring engine reads the finished work. The engine grades eight things: format, hook, value, clarity, tone, call to action, fit, and density. Its baseline comes from my own best posts, drawn from a corpus of nearly 500 of my X posts that re-syncs twice a week. A weak hook shows up right away. Whether I stand behind the words never shows up at all.
Rabbi Dovid Tzvi Hoffmann, who taught for a time at Hirsch's school in Frankfurt before leading the rabbinical seminary in Berlin, was once asked about a new invention. Someone had built a clock that could be set before Shabbat to open a gas valve at a fixed hour, turning a small pilot flame into a full one under a pot. Hoffmann worked through the question on its own terms and wrote, in his responsa Melamed LeHo'il (III:58), "In the end, I went back over every side and found no reason to forbid it."
Then he drew a line. He wouldn't permit using the device to cook a raw dish from the start. He permitted it for warming food that was already fully cooked.
His reasoning was halachic, not literary, and I don't want to stretch it into something it isn't. But the shape of his answer stuck with me. He didn't dismiss the new tool and he didn't surrender to it. He studied it closely and was specific about which part of the work it was allowed to touch. For me, the cooking is the writing. Tools can prepare the kitchen and taste the dish afterward. The cooking stays mine.
Sincerity is the part I can't delegate
The philosopher Bernard Williams, in Truth and Truthfulness, names two virtues of truth. He calls them Accuracy and Sincerity: "you do the best you can to acquire true beliefs, and what you say reveals what you believe."
That distinction maps neatly onto my process. The brief helps with accuracy. It points me to sources I can check and flags claims that need support. The score helps me see whether the piece will reach anyone. Neither can help with sincerity, because sincerity is the match between the words and the person behind them. It's Maimonides' heart and lips again, eight centuries apart.
There's a danger on the scoring side, too. The anthropologist Marilyn Strathern gave Goodhart's law its best known form: "When a measure becomes a target, it ceases to be a good measure." If I start writing toward whatever the engine rewards, I'll lose the reason I wanted to write the piece in the first place. So I treat the score as a reason to look again, never as a verdict. When I disagree with it, I have to be able to say why, and then I keep the sentence.
In the name of the one who said it
Pirkei Avot 6:6 lists forty-eight ways Torah is acquired. The last one is saying something in the name of the person who said it, "everyone who says a thing in the name of him who said it, brings deliverance into the world." The Talmud in Megillah 15a derives it from Esther, who told the king about the plot against him "in the name of Mordechai."
When you name where an idea came from, the reader can weigh it for what it is.
Hoffmann understood this about his own scholarship. In the preface to his Leviticus commentary, written in Berlin in 1904, he admits openly that his faith commitments shape his conclusions about who wrote the Torah. He says he tried to make only arguments that someone standing elsewhere could also recognize as valid. And if his defense seems weak in places, he writes, put it down to his own inadequacy.
He says where he stands, shows how he argued, and takes the blame for any weak spots himself. That's what I'd like a disclosure to do.
So here's mine, in plain terms. An agent helps me gather context before I write. The argument and the sentences in between are mine, and if something here doesn't hold up, that's on me, not on the tools.