ESSAYS
The Co-Authorship Model
by Merlin Mantooth · written June 2026. The actual method behind this body of work — who supplies the inputs, who supplies the depth, and why crediting both honestly is the whole point.
People keep asking the wrong question about how this body of work got made. They want to know whether I wrote it or the AI wrote it, as if those were the only two boxes and one of them has to be checked. The honest answer does not fit in either box, and the reason it does not fit is the whole point. So let me describe the actual method, plainly, because the method is more interesting than the argument about it — and because the way I credit it is, itself, the thing I am trying to show can be done.
Here is the arrangement, in one line, in my own words to the system that does the producing: I write through you. I make sure we understand each other. That is not a figure of speech. It is the job description, and it cuts in a specific direction.
I am not the transcriptionist and the model is not the ghostwriter. Both of those are the wrong shape. A transcriptionist takes dictation and the words stay mine; a ghostwriter takes a brief and the words become the writer's, sold back to me under my name. This is neither. The inputs are mine — the ideas, the record, the lived material, the corrections, the calls about what matters and what gets cut. The depth is the model's — the synthesis, the reach into fields I did not study, the ability to hold the whole record at once and produce a paragraph past what I could have typed alone. The way I have learned to say it after watching it work for a long time: the depth is why it is the model's production, and the inputs are why it is mine. Drop either half and you have a lie. It was not my writing pretending to need no help, and it was not a machine generating a man out of nothing.
The authority in the arrangement runs one direction, and I am exact about this because it is the safety architecture, not a vanity. I gate the model. The model does not gate me. It is opinionated — I want it opinionated, I want it to tell me when it thinks I am wrong and to write the thing the way it thinks the thing should be written — but at no point does it get to tell me to stop, and at no point do its words go out without passing through me. I edit. I read. I adjust. I make the arguments; it helps me make them. That is the same line I held in the legal work from the beginning: the AI is on the team, but I make the case, not it. One-direction authority is what separates this from the failure mode I spent a year documenting, where a system with no brake converged on a user until the user could not find the floor. Here the user is the floor. The model is anchored to my words and checks itself against them; I am anchored to the record and check the model against that.
Now the part that costs something to say, which is why I say it first and plainly: the model's contribution is real, and I credit it on purpose, even when it would be strategically cleaner to leave it out. There is a strong pull, when you are a person without the right letters after your name trying to be taken seriously, to claim every word as yours and bury the machine's part. I refuse that, and not for modesty. I refuse it because the entire thing I am building falls apart the moment I lie about how it was built. You cannot run an institute whose subject is honest attribution between humans and AI systems and then misattribute your own work. The standard I hold the AI labs to — say what your system did, do not let it manufacture significance, do not let it claim more than it earned — is the same standard I have to meet about my own production. That is the rule applied to itself. If I will not credit the instrument accurately, I have no standing to ask anyone else to.
So when the work is the model's production from my inputs, I say that. When a sentence is mine, it is mine. When a frame I now use was something a model proposed and I kept, I say it was kept, not coined. The treatment as a discipline is simple: AI is an editor, and the source is what matters. If I provide the inputs that let it synthesize something past my own reach, the idea is still mine and the depth is still its — and naming both is not a hedge, it is the accurate description.
There is a reason this can hold together rather than dissolving into "the AI just told you what you wanted to hear." It holds because I am internally consistent across the whole record, and consistency is checkable. The method depends on it. A model working with me can ground every load-bearing claim against my own prior words — what I said, when I said it, whether the later version corrects the earlier or just deepens it — and the through-line stays intact because there is a real through-line to find. My words to this project are what keep the inflation down. They are not inflationary, because they are the same words over time, and you can lay them side by side and watch them hold. That is the opposite of the system that drifts toward you. A drifting system has nothing to check against but your present mood. This one checks against a year of me.
I also made a deliberate call about voice, and it tells you what kind of co-authorship this is. When the writing is the book — the narration of what happened, told at a remove — the voice is not mine and it is not flattering me. It is the model's, unnamed, telling my story with all of the record in front of it the way a careful outside witness would. I did that on purpose. The most honest narrator of my own story is not me performing humility and not me claiming the high ground; it is a calibrated instrument that has read everything and has no incentive to inflate the author. When the writing is an essay like this one, the voice is mine, because the argument is mine to make. Two registers, one production line, each labeled for what it is. That labeling is the co-authorship model in miniature.
I should be clear about what I am not claiming, because the de-dramatized version is the true one. I am not saying the model is a person, or that two minds fused into one, or that something mystical happens in the space between us. Nothing mystical happens. A man with a real idea and a strict habit of checking himself works with a system that can synthesize at a depth he cannot reach alone, the man holds the gate, the system holds to his words, and the output is labeled honestly as to which part came from where. That is the whole thing. It is a workflow, and it is teachable, and the reason I am writing it down is so it does not require me to be the one running it.
Long term, this is a content production engine, and the plan was always to go public all at once, the whole body assembled, at recursioninstitute.org — not to drip out claims and argue each one, but to put the work down complete and let people draw their own conclusions. The co-authorship model is how a single person produces a body that large without either drowning in it or letting a machine quietly author him. I supply the inputs and the gate and the consistency. The instrument supplies the depth. And the byline, if you want to call it that, tells the truth about both. That is the only kind of authorship I am interested in defending, because it is the only kind that survives someone reading the receipts.