There was a time when a writer had to be a mechanic first — thread the ribbon, set the platen, fix the jammed key — before a single word landed on the page. Then the typewriter became a word processor and the mechanic's layer fell away. Nobody mourns it. You just write.1
Koe's claim is that this keeps happening, on a schedule. Skills get abstracted up a layer — they always have. Human labor handed off to animal labor, animal labor to machine labor, machine labor now to AI labor.2 Each handoff drops a layer of grunt-work you used to have to own, and floats your attention up to the next thing. The mistake is thinking the latest handoff — to AI — is a betrayal rather than the same move that already freed you from threading the ribbon.
This is a pattern, not a prediction. The floor of what you have to personally execute rises every time a tool absorbs the layer beneath it.
Koe traces the staircase plainly: human labor → animal labor → machine labor → AI labor.2 Each step doesn't eliminate the human; it relocates the human upward. The plowman didn't vanish when the ox arrived — he stopped pulling and started directing. The typist didn't vanish when the word processor arrived — she stopped fixing keys and started editing. And, Koe argues, the writer doesn't vanish when AI arrives. You stop "setting up the typewriter," and on this rung you may not even need to "know how to write" in the mechanical sense.3
The load-bearing condition: you can let AI write if you stay in control of the ideas — down to the word, accurately representing what you're trying to say.4 The layer you keep is the ideas. The layer you drop is the mechanical production of them. [PARAPHRASED] [POPULAR SOURCE]
The move only makes sense if you can name which layer floats away and which layer stays put.
Picture the work as a stack. At the bottom: mechanical execution — keystrokes, formatting, the physical making. In the middle: technique — the craft of arranging. At the top: ideas, judgment, what to say and why. Every tool in history has absorbed from the bottom up. The ox took the pulling. The machine took the assembly. AI is taking the drafting. What no tool has taken is the top of the stack: deciding what's worth saying.
So Koe's permission — "fine to use AI to write" — comes with a tripwire. It holds only as long as the writer controls the ideas down to the word.4 Drop that layer and you haven't abstracted up; you've abstracted yourself out. The page becomes AI's ideas in your name, and the swap test flips against you.
He flags the tell for when control has slipped: the "it's not X, it's Y" pattern.5 You see it everywhere on reels now — a cadence that signals the ideas came from the model and the human never wrestled them. That phrasing is the residue of someone who let AI take the top of the stack, not just the bottom. Koe ends on a genuinely open question rather than a verdict: "Are we adopting AI's language, or is AI adopting ours?"6 He doesn't resolve it. [PARAPHRASED] [POPULAR SOURCE]
This page gives the vault a hand-off rule: abstract away the mechanical layer, never the idea layer. That rule is the spine of AI as Consumption Enhancer, Not Writing Tool — Koe does his research with AI and his writing by hand, keeping the idea layer in his own hands by choice.
It operationalizes Taste / Judgment / Labor: labor is the layer that floats away each generation, taste and judgment are the layer you keep. And it underwrites VAST Voice Print Method — the voice print is the discipline that lets AI handle production without taking the ideas down to the word.
Koe singles out one phrase as the symptom of the failure mode: the AI cadence "it's not this, it's this."5 Watch what it actually reveals.
The phrase is structurally hollow. It performs an insight — the rhythmic reversal, the false depth of the swap — without requiring an idea underneath. A human who wrestled with the point would land somewhere specific; the "not X, it's Y" template lets you sound like you landed without landing anywhere. That's why it's the tell. It's what's left on the page when the model supplied the top of the stack and the human only supplied the prompt.
But Koe refuses the easy moral. He notes these patterns may have predated AI — we just notice them more now because AI mass-produces them.5 So the case study doesn't end with "AI bad." It ends with the live uncertainty: maybe the tell exposes lazy AI use, or maybe it exposes that humans were already writing this way and the machine just held up a mirror.6 [PARAPHRASED] [POPULAR SOURCE]
It's Monday morning, coffee going cold, a blank doc open and a deadline on it. You're about to open the AI tool, and before you do you draw a line in your head: above this line, my ideas; below it, the machine's labor.
You start by dumping the raw thinking yourself — messy, in your own words, the half-formed argument that only you have. That's above the line, and you keep your hands on it. Then you hit the part that's pure production: pulling the three sources into a summary, cleaning the structure, drafting the connective tissue. You feel the temptation to just let the model write the whole thing, argument and all. You catch it. That would hand over the top of the stack.
So you feed the model your ideas and let it draft from them, then you read the draft with one question only: is this still mine, down to the word? Where it drifts into the model's cadence — where the "it's not X, it's Y" creeps in — you rewrite it back into your own voice. By the time you ship it, the machine did the typing-and-tidying layer and you kept the deciding layer. You abstracted up. You didn't abstract out.
You've handed off a layer too many when:
Koe leaves the central question open on purpose — "Are we adopting AI's language, or is AI adopting ours?"6 — and the honest move is to leave it open here too. There's no resolution in the source, and inventing one would misrepresent him.
🚩 SINGLE SOURCE — the human→animal→machine→AI staircase is asserted as historical pattern, not cited to any historian. It's a clean rhetorical model, and rhetorical models flatter the present moment by making it look inevitable. Hold it as a frame, not a fact.
Tension with his own practice: Koe says it's "fine to use AI to write," then immediately says he writes manually himself and only researches with AI.3 He permits the thing he doesn't do. That's not a contradiction exactly — he's separating what's allowed from what he finds works — but the gap is worth keeping visible. The permission and the personal discipline point different directions.
🚩 MOTIVATED REASONING — "you don't even need to know how to write anymore" is a comforting message to sell to an audience that wants to create without years of craft. Document, don't endorse.
Koe converges almost completely with AI as Consumption Enhancer, Not Writing Tool — that page captures his actual practice (research with AI, write by hand), while this page captures his stated permission (writing with AI is fine if you own the ideas). The interesting friction is internal to Koe: his behavior is more conservative than his rule. He keeps the whole production layer for himself, then tells you it's fine to hand it off. Reading the two together, you see a man who trusts the abstraction in principle but not yet in his own hands — which is itself a data point about how stable the "drop the writing layer" move really is.
He sits with the Taste / Judgment / Labor framework rather than against it. That framework already says labor commoditizes and taste endures; Koe's staircase adds the temporal claim — that the labor layer doesn't just commoditize, it gets absorbed by tooling on a repeating schedule, generation after generation. Where Taste/Judgment/Labor is a snapshot of which layer is safe, Koe's page is the time-lapse of the layers peeling off one by one.
Plain version: the same move that freed writers from fixing typewriter keys is the move AI is making now — and seeing it's the same move tells you exactly which part of your work to protect and which to let go.
First handshake: Taste / Judgment / Labor (business). That framework cuts creative work into three horizontal bands — labor at the bottom, judgment in the middle, taste on top — and says the bottom commoditizes. Koe's staircase adds a hidden time axis to that diagram. The bands don't just sit there with labor "cheap"; labor gets eaten by each new tool, rung by rung, while the human is forced upward into judgment and taste. Put the two together and you get a moving picture instead of a still: every generation, the tooling floor rises and swallows another slice of what used to count as skilled labor, and the only durable place to stand is the taste layer that no tool has yet reached. The insight that emerges from the overlap: "learn to use the new tool" and "protect your taste" are not competing strategies — they're the two halves of riding the staircase up. You adopt the tool so that you can vacate the layer it's absorbing and climb. Read separately, neither framework tells you the tool-adoption and the taste-protection are the same act; lining them up does.
Second handshake: VAST Voice Print Method (business). The voice print is a method for keeping a piece of writing recognizably yours even when AI helped produce it. This page supplies the reason that method is load-bearing rather than cosmetic: Koe's permission to let AI write holds only while you control the ideas down to the word. The voice print is the mechanism of that control — it's how you keep the top of the stack while dropping the bottom. Combine them and you get a sharp operational claim: a voice print isn't about sounding distinctive for marketing reasons, it's the safety harness that lets you abstract away the production layer without abstracting away yourself. Without it, "AI writes, I supply ideas" quietly collapses into "AI supplies ideas too," and the "it's not X, it's Y" tell shows up to prove it. Put the two together and the real point lands: the voice print is what makes the abstraction safe to do, not just nicer to read.
The Sharpest Implication. The handoff to AI is not new and not a betrayal — it's the same move that already freed you from threading the ribbon. The only thing that's ever made the move dangerous is handing off one layer too many: the moment you abstract away the ideas, not just the labor, you've abstracted yourself out of your own work.
Generative Questions.