Supervised hybridization in writing
I'm writing this sentence, and then I'm not.
That contradiction is the whole point of what follows, so let me be honest about it from the first line rather than burying it somewhere at the bottom the way I sometimes do when I'm not sure how a confession will land. This piece was drafted by Claude, under my direction, fed by my own writing, my own frameworks and the conversations we've been having about this project. I supervised every choice that mattered. I did not type every word. This is an exception, made for this specific occasion, and I want to think through it with you rather than just announce it and move on.
Let's call the thing itself supervised hybridization.
Not because I'm fond of clinical-sounding terms, but because the phrase actually does some work. Hybridization because two different kinds of cognition are producing one text: mine, shaped by years of thinking in three languages and a particular obsession with agency, and a model's, shaped by an almost incomprehensible amount of human writing compressed into weights. Supervised because none of that hybrid output reaches you without passing through my judgment first, the same way a translation doesn't reach a reader without passing through the translator's.
I keep coming back to translation as the closest analogy, and it's not an accident. Anyone who has translated a text seriously knows the discomfort of this position: you are not the author, but you are also not a passive conduit. You make thousands of small decisions the original writer never had to make, and the final text carries your fingerprints even though the ideas aren't yours to claim. Translators have lived with this exact ambiguity for centuries and mostly made peace with it. Writers working with language models are just now being asked to.
Where this fits inside this project
If you've been reading along, you know the project runs on two frameworks. One is pedagogical: metacognition through language learning, the whole architecture I built around intentional language and the Spanish-speaking persona I described in an older newsletter. The other is a translation workflow, meant to let me work closely with authors and institutions I can have genuine conversations with, before those conversations become something more public, like a campaign or a piece of shared work.
Supervised hybridization sits right at the seam between the two.
On one side, it's a metacognitive exercise. Directing a model to write in my voice forces me to articulate what my voice actually is, which sentences are mine and which ones only sound like they could be, where the framework ends and the personality begins. I've asked my students to do something similar in Spanish: build a persona out of language, then examine it. I hadn't expected to run the same exercise on myself, in English, with a machine instead of a mirror.
On the other side, it's a preview of the translation workflow itself. If I can supervise a model closely enough to produce something that sounds like me, I can supervise it closely enough to carry someone else's ideas across a language boundary without losing what made them worth translating in the first place. The discipline is the same discipline. Only the source changes.
The uncomfortable part
I'd be lying if I said this felt entirely comfortable to publish. There's a small, insistent voice asking whether a reader who liked my earlier newsletters liked me, specifically, or liked a style that turns out to be reproducible by anyone with access to the same conversations I've had with Claude. I don't have a tidy answer. What I can tell you is that the ideas in this piece, the framework, the vocabulary, the references, the way I connect a translator's discomfort to a language learner's persona, all of that came from months of my own writing and thinking, fed back into the system as material. Nothing here is invented on my behalf. It's arranged on my behalf, closely, under my direction, and that distinction matters more to me than I expected it to.
Extended cognition, not outsourced cognition
Andy Clark's idea of the extended mind is useful here: the notebook, the calculator, the search engine, none of them think for you, but all of them become part of the loop your thinking runs through. A language model under close supervision is not a different category of tool, it's a much more articulate one, and articulate tools change the shape of what gets produced without necessarily changing who's producing it.
Michael Levin's thoughts-as-thinkers idea, which I brought up in another newsletter, applies here too, oddly enough. If thoughts already act somewhat independently inside a single cognitive system, then a supervised model isn't introducing some radical new kind of otherness into the writing process. It's making visible a kind of otherness that was arguably already there, just distributed differently. The dragon of AGI I mentioned back in December hasn't gone anywhere. It's just started drafting.
What I'm asking myself, and now you
I don't think this becomes the default mode for this newsletter. The whole reason this project exists is to defend a very specific, very human kind of agency, and I'm not interested in quietly automating my way out of the thing I keep asking my students to develop. But I also don't think I can honestly pretend this kind of hybrid writing is a fringe curiosity anymore. It's already happening across newsletters, marketing copy, even literary fiction, mostly without the disclosure I just gave you.
So here's the actual question I want to sit with, and I'd genuinely like to know where you land on it:
If supervised hybridization becomes a normal part of how most writers work within the next few years, closely directed, transparently disclosed, fed by their own material, does that change what you expect from a byline?
Does knowing the process change how you read the last few paragraphs, now that you know?
I don't have a settled position yet, which is exactly why I wanted to publish this one openly instead of quietly. Reply and tell me how this sits with you. I read everything.
See you soon,
Javier