You've read these. The grammar is fine. The structure is competent. Yet something is wrong, and within a few paragraphs you can name what — every sentence sits at the same register, the same confidence level, the same emotional temperature. The writer never switches. The voice is mode-locked.
That uniformity is the AI tell. Michael Deen names it directly: "If you confine yourself to a single mode of expression, it triggers the uncanny valley effect. It feels inhuman, as if the writer isn't sensitive enough to bend their voice to suit the context."1 In a different chapter: "If you hide behind abstractions, your readers won't trust you. They might even wonder if ChatGPT wrote it."1
The AI argument runs through the entire Essay Architecture framework. Not as side commentary. As structural motivation. The whole framework is, in part, an argument about why essay craft matters now, and the argument is: human writing carries a moat that AI cannot easily cross. The framework names the moat operationally.
Three distinct claims about the AI threat run across the framework:
The rumors-of-ancestors claim (Material): Personal Experience is irreplicable. "A large language model can read trillions of words, but it won't ever learn the rumors of my ancestors."1 Your specific family history, your particular life events, your unrepeatable trajectory through specific contexts — this is a private dataset that no model has access to.
The uncanny-valley claim (Spirit): Multi-axis modulation is hard for AI. "If you confine yourself to a single mode of expression, it triggers the uncanny valley effect."1 Conversely: organic mode-switching across register, confidence, valence, distance signals human-rendered prose.
The flat-sentences claim (Wang's adjacent argument, converging with Deen's framework): AI defaults to flat prose. Wang reports that AI writing produces "generally super flat sentences" and is explicitly verboten in his own writing process.2
Three different attack surfaces. Three different aspects of essay craft that AI struggles with. Together they form a structural defense.
Deen doesn't use the word "moat" directly. The structure is implicit. Stack the claims:
The essay form, as Deen describes it, is precisely the form where all three matter:
The form is, by design, the form AI struggles to imitate. Not because anyone designed it that way — but because the form happens to be optimized for cognitive properties that humans have and current AI doesn't.
The implication: writers who master Deen's framework are operating inside the moat. Their work is harder for AI to replace than the work of writers who don't modulate, don't draw on Experience, don't trust subtext.
Material chapter (1): Family-rumors moat. The whole logic of personal Experience as the irreplicable layer. "My life is a reservoir of Material that nobody else has access to. This is my moat."1
Experience sub-pattern (1.1): "If you hide behind abstractions, your readers won't trust you. They might even wonder if ChatGPT wrote it."1
Spirit chapter (7): Multi-axis modulation as human signature. The whole Tone framework (register/confidence/valence/distance) is explicitly engineered as anti-mode-lock.
Tone sub-pattern (7.1): "Sophisticated tonal control breeds trust, and AI models can't replicate this (yet)."1 The parenthetical (yet) matters — Deen is hedging on whether the moat will hold.
Voice meta-essay: "As the Internet gets crowded and AI advances, less writers will doubt the importance of voice; the real debate is on where voice comes from."1 The frame: AI raises the importance of voice precisely because voice is what AI struggles to reproduce.
Throughout the framework: every rubric's 5/5 bar involves properties AI doesn't currently achieve — "shape-shifter," "earworm meme-status," "holy shit fractal compression," "making inaction feel impossible."
The AI-defense argument makes a falsifiable claim: as AI models gain capacity to modulate across multiple axes simultaneously, the moat narrows. The 9-axis framework becomes more important, not less, in proportion to AI capability gain.
This is testable. You could measure:
So far, the moat seems to hold. AI prose reads as mode-locked even when it's technically correct. But Deen's (yet) signals he knows the moat is narrowing. The framework's implicit claim: writers who train modulation get further inside the moat as the moat narrows.
The strategic implication: writers who optimize for consistency (brand voice, recognizable signature, repeatable formula) are easier to replace by AI than writers who optimize for modulation. Brand-voice is the most replicable kind of voice for a model to imitate. Modulation is the hardest.
This is uncomfortable because the writing-economy incentive structure rewards brand-voice. Substack newsletter writers, podcast personalities, columnist contracts — all push toward stable signature, predictable voice, recognizable take. The market structure pulls writers toward AI-replaceable form precisely as the AI capability is rising.
Deen's framework is, in this sense, anti-market. The skill it cultivates is the opposite of what the market currently rewards. The bet: this will invert. Mode-locked writers will be replaced first; modulation-rich writers will survive longest.
Stack the framework as defense moves:
Lead with Experience that no model has access to. Family stories. Specific places. Specific moments. The dataset that's only yours.
Modulate across the four Tone modes within each essay. Register shifts. Confidence shifts. Valence shifts. Distance shifts. Don't write entire essays at a fixed point on any axis.
Deploy Subtext rather than over-explanation. AI defaults to over-explanation because over-explanation is safer. Trust your reader. Imply more than you say.
Use Imagery with specificity. Alpenglow. Eighteen inches. The rare or unexpected word that lands. AI defaults to common vocabulary in standard combinations.
Build a Motif that accumulates meaning across the essay. AI can mention an image multiple times; AI struggles to evolve the image's meaning across instances in ways that load earlier context.
Engineer Sound at the high-leverage moments. Title, Hook, Finale. AI can produce competent prose but struggles with the phonetic compression that produces earworms.
Hide-behind-abstractions failure: You avoid personal Experience because it feels indulgent. Your essay reads as anyone-could-have-written-it. The fix: lead with the texture only you can deliver.
Brand-voice failure: You've trained yourself into a stable recognizable voice. You're a moving target for AI imitation. The fix: deliberately deploy your axis-opposites at least once per essay.
Over-explanation failure: You spell every implication out. AI does this perfectly. The fix: dedicated Subtext pass. Trust your reader.
Generic-imagery failure: Your concrete details could appear in any essay on the topic. The fix: McPhee's circle-stale-words protocol. Replace the generic with the specific.
Predictable rhythm failure: Sentence-length sits in a narrow band. AI mimics this easily. The fix: deliberately vary. Some sentences 7x shorter or longer than your average.
Audit Experience density: How much of your essay's value comes from your specific dataset (family, place, era, profession) vs. from facts anyone could look up? Increase the former.
Audit modulation: For each of the four Tone modes, mark your default. Mark where you deployed the opposite. If you didn't, add at least one deployment per axis.
Audit Subtext: Find the A→B→C→D causal chains. Collapse them. Force the reader to fill B and C.
Audit Imagery specificity: Circle every stale or generic word. Replace with something specific. Generic produces AI-comparable output; specific does not.
Audit Motif evolution: Does your recurring image accumulate meaning across instances? Or does it just appear repeatedly with the same reference? Engineer the evolution.
Where Deen aligns with Dan Wang on AI-as-flat-prose: Wang's ai-as-consumption-enhancer-not-writing-tool page explicitly bans AI from his writing process because AI produces "generally super flat sentences."2 Deen's framework explains why — flat sentences are mode-locked across the four Tone axes. The convergence: both authors locate AI's writing weakness in modulation capacity, not in factual knowledge or grammatical correctness. The split: Wang's stance is practical (he doesn't use AI for writing); Deen's stance is theoretical (the framework explains why AI struggles). Together they make a complete case.
Where Deen complicates the AI-replacement-narrative: standard discourse on AI and writing is either alarmist (AI will replace all writers) or dismissive (AI is just a tool). Deen takes a third position: AI raises the bar on what writing has to be to be human. The framework isn't a defense against AI extinction; it's a description of the kind of writing that survives AI's improvement. The split with both alarmist and dismissive positions: Deen treats AI as an opportunity to refine craft, not as either threat or non-issue.
Essay-craft-as-AI-defense reaches into adjacent vault territory at three specific points.
AI Collaboration — AI as Consumption Enhancer, Not Writing Tool (Wang): Wang and Deen converge on the structural argument — AI is bad at the multi-axis modulation that human voice deploys. Wang's evidence is experiential (his students using ChatGPT produce flat outputs); Deen's evidence is operational (mode-locked prose triggers uncanny valley). Together they reframe the threat: as models improve at multi-axis modulation, the moat narrows. The 9-axis framework becomes the testable definition of what survives capability gain. The handshake reveals: the AI defense is not magical; it's measurable. The framework provides the metrics.
Behavioral Mechanics — Uncanny Valley Detection as Trust Signal: The uncanny valley phenomenon (Mori 1970) describes the dip in human response when a thing is almost but not quite human-like. Reader response to mode-locked AI prose follows the same psychology — readers detect almost-but-not-quite-human signals and respond with distrust. Deen's modulation-breeds-trust claim is the writing-craft version of the uncanny-valley principle. The structural identity: the cognitive system has fine-grained detectors for human-vs-not-quite-human signals, and those detectors operate on rhetorical-modulation cues. The implication: the uncanny valley moves as AI improves. The writing that today reads as clearly human will gradually become AI-imitable; the moat narrows over time but doesn't disappear because human modulation has fractal complexity that compounds.
Cross-Domain — Private Dataset as Economic Moat: The standard moat analysis in AI strategy holds that whoever has the proprietary data has the durable advantage. Open data is commoditized; proprietary data is moat. Deen's family-rumors argument operationalizes this for individual writers. Your personal Experience is your proprietary dataset. Deen's framework treats individual writers as small-scale AI competitors. The strategic logic for a multi-billion-dollar AI company and for a solo essayist is identical — your moat is the data nobody else has. For the company that's a private dataset; for the writer it's the rumors of your ancestors.
Eastern Spirituality — Svādhyāya: Self-Study as Practice: One of the niyamas in the eight-limb yoga system, svādhyāya means self-study — observing one's own thought, sensation, and reaction with sustained attention. The contemplative tradition's claim: the practitioner's own consciousness is irreplicable territory that no external study can substitute for. Deen's family-rumors argument is the writing-craft version of this. The dharmic tradition has known for millennia that the writer's most defensible asset is their own attention to their own life. Modern AI strategy has rediscovered this finding under the name private dataset. The handshake reveals: the strongest moat available to any human producer in the AI period is the irreplicable consciousness only they can document. This was always true. The AI threat just made it visible.
Synthesis across the four connections: each domain has independently identified the same fact about the AI period — what survives is what cannot be averaged. Wang's flat-prose observation, Mori's uncanny-valley research, AI moat theory, and dharmic svādhyāya converge on the same operational claim: the producer who can document something irreplicable about themselves has durable advantage; the producer who optimizes for average-of-genre is replaceable. Deen's framework names the writing-craft case. The broader pattern applies across all human production. Writers, artists, founders, researchers — the strategic logic is uniform. Lead with what no model has access to. Refuse to compete on what models also produce. The moat is consciousness, specifically yours, specifically frozen.
The Sharpest Implication: If essay craft is structurally optimized as defense against AI homogenization, then most contemporary writing pedagogy is teaching the wrong thing. Brand-voice optimization, SEO-driven structure, consistency-as-virtue — all are exactly the properties that make a writer most replaceable. The market trains writers into AI-vulnerability. Deen's framework trains writers into AI-resistance. The strategic implication for any writer building a career: optimize for modulation, Experience, and Subtext. Optimize against consistency. The shortest path to durable advantage is also the path the market doesn't currently reward. This is uncomfortable. It implies that the writers who get most-rewarded by the current market are also most-replaceable by the next-generation models, and the writers who'd be most-durable are getting less reward today.
Generative Questions:
Maria Popova adds the deepest version of the AI-defense claim the page contains.2 The standard argument is that AI cannot replicate individual experience. Popova's argument goes one layer deeper. AI cannot suffer. The relevant capacity is not access to a particular life but the encounter with impossibility — the moment when a writer who has been working on a piece for months runs into the wall of what cannot be said and has to go through that wall to produce the next sentence. The going-through-the-wall is what makes the resulting prose carry weight. AI does not encounter walls. AI does not suffer the encounter. Therefore AI prose, however polished, lacks the specific weight the suffered-through prose carries.
Popova's exact claim: AI will never know what it's like to collide with its own impossibility. The phrasing is precise. The collision is the requirement. The impossibility is the wall. The knowing-what-it's-like is the irreplicable element. Deen's frame identifies the genres of irreplicable material (Experience, Subtext, Modulation). Popova identifies the operation that makes the irreplicable material weighty — the writer's body-level encounter with the limits of what can be rendered, and the work that happens at those limits.
The implication for the page: the moat is not only what the writer has access to but what the writer has done to encounter what cannot be rendered. The writer who has not done the suffering produces lighter prose even when the content is technically present. The reader detects the lightness. The AI prose is light in this specific way. The writer who has suffered the collision produces something that even AI prose informed by the writer's content cannot reach.
The frame is structurally connected to Karlsson's depth-moments and Tucker's wound-to-scar discipline. All three are describing variants of the writer being changed by the encounter with what cannot be said. The changed writer produces different prose than the unchanged writer. The change is operationally inaccessible to AI. The moat is the change, not the content.