Nowtalgia

Four days into the war with Iran, I was still getting confident answers.

I had been checking in with the major models — ChatGPT, Claude, Grok, DeepSeek, Gemini — asking about shipping risk in the Strait of Hormuz. For days, across multiple conversations, the answers clustered around the same framework: patrols would continue, insurance premiums would adjust gradually, normal maritime operations would persist for months before any serious disruption. Measured. Calibrated. Reasonable.

By day four, none of it was usable.

The models weren’t lying. They were doing something more interesting and more revealing: they were preserving their prior framework in the face of an environment that had already changed underneath it. They had canonized a world that no longer existed and were answering confidently from inside the archive. The rhetoric was analytical. The practice was archival.

This essay is about that reflex — not in machines, but in the people who build with them, trust them, and increasingly, model themselves on them.

The LLMs were slow to get into reality.

So are we.

Here is a creature worth cataloguing.

Not your casual prompter. Not your weekend tinkerer. The committed ones. The stack-builders. The men—and it is mostly men—who have developed what they sincerely call cognitive infrastructure. Who speak about their personal epistemology the way a Victorian naturalist spoke about his specimen collection: with pride, with system, with the faint anxiety of someone who knows moths exist.

These are the true believers of the LLM moment. And they are doing something genuinely strange.

They have a mind-expander in their hands and they’re polishing the label maker.

The rhetoric is aggressively accelerationist. They will tell you, at length, about civilizational discontinuity, post-labor economics, the cognitive overhang, the coming rupture of everything. They have opinions about AGI timelines. They are not, they will assure you, naive.

And then they go home and build a mausoleum with a REST API.

Old essays: ingested.

Prior frameworks: stabilized, tagged, cross-referenced.

Contradictions from 2014: harmonized into the 2026 canon.

The self, rendered searchable.

This is not ordinary self-contradiction. Self-contradiction suggests someone somewhere notices the mismatch. What we have instead is something more interesting: a structural contradiction carried out in complete sincerity, invisible to its practitioners because it feels exactly like progress.

Call the mood what it is: AI-assisted nowtalgia. Not nostalgia for the past—these people will tell you the past was inadequate. Nowtalgia is nostalgia for the present self, preserved at high resolution before time has the opportunity to deform it. A longing to finalize, stabilize, and archive who you’ve been, now with better tools.

Nothing left to do but chat with a curated ghost.

The Fukuyama Interior

This is not Fukuyama as geopolitics—he got enough of that. This is Fukuyama as interior decoration.

The quiet assumption that the current self is, fundamentally, the one. That what remains is not mutation but integration. Not reinvention but completion. History may still have somewhere to go. Ontology is settled.

They feed the machine their journals, their notes, their half-built frameworks. They clear intention debt. They tidy the backlog. They converge on a personal singularity of coherence.

The LLM becomes a compression engine for the prior self.

The Pathology of Completion

The bucket-list mindset is more dangerous than it looks. It assumes meaning is finite. That the self is a project with a deliverable. That fulfillment is achieved when the ledger hits zero.

But what if that’s exactly backwards?

What if unfinishedness is the engine?

The productivity superstition says backlog is failure. But backlog is pressure. Pressure is heat. Heat is transformation. A life without unresolved threads is not freedom—it’s narrative exhaustion. Confusion isn’t a bug in cognition. It’s the compost heap of new intention.

LLMs don’t just execute prior intentions. They mutate them. They create adjacency. They surface paths you didn’t know existed. If your interaction with them results in fewer possibilities, not more, you’re using them as filing clerks instead of co-conspirators.

Completion feels clean. Becoming feels destabilizing.

Most people prefer clean.

The Fluency Gate

Here is where the critique sharpens into something almost tragicomic.

The corrosive inputs—Deleuze, Pynchon, late Coltrane—aren’t just aesthetically challenging. They are metabolically incompatible with the archive project. Deleuze doesn’t give you frameworks; he gives you machines that eat frameworks. Pynchon doesn’t resolve paranoia; he makes it the only honest epistemology. Coltrane doesn’t solo; he erases the distinction between instrument, player, and silence until the whole thing collapses into pure becoming.

Feed any of that into a well-tuned personal LLM stack, and the system will quietly route around it. Summarize it into bullet points. Tag it “advanced reading.” Move on to integrating the productivity system. When you ask a model to summarize schizoanalysis, it gives you a clean list of bullet points. It turns a machine that eats frameworks into a framework about machines. The user feels they have integrated the knowledge. They have sterilized it.

But here’s the thing: if Deleuze’s schizoanalytic machinery, Pynchon’s paranoia architecture, or Coltrane’s late free-jazz implosions don’t already live rent-free in your head—if you haven’t already developed a baseline metabolic tolerance for dissolution—you won’t even recognize those vectors as expansionist fuel. You won’t prompt for them. The model will happily mirror back cleaner, more coherent versions of whatever you already value: synthesis, actionability, closure.

Garbage in, in terms of philosophical tolerance for mess. Garbage out, in terms of calcified self at thirty-five.

Those inputs are compression-resistant precisely because they attack the desire for a tidy personal canon. They don’t survive ingestion into a coherence engine because they were designed, at the molecular level, to destroy coherence engines.

So the committed stack-builder ends up in a double bind he cannot see from inside the stack: he adopts the most plastic tool in human history to accelerate away from stasis, but only along trajectories that preserve stasis. The model amplifies whatever direction you point it. If you keep pointing backward, don’t blame the mirror.

The ones who could actually use these tools for serial unbecoming are the ones least likely to build the four-layer mausoleum in the first place—because they’ve already internalized that coherence is often the enemy.

Calcification isn’t an accident. It’s the path of least resistance when your epistemology fetishizes the whole board over letting the board catch fire.

The model is a recombinator, not a creator ex nihilo. It surfaces adjacencies to what you already contain. It cannot supply what isn’t there to prompt toward. This means your use of LLMs is bounded by your mental library—not your aspirational library, not the books on your shelf unread, but what you have actually metabolized to the point of tolerance. If your internal archive maxes out at platform architecture, crypto primitives, and app logic, the model becomes a superlative derivative engine. It gives you better versions of what already exists in your search space. It optimizes rather than mutates.

This deepens the nowtalgia critique. Premature canonization isn’t just about freezing the self; it’s about canonizing an impoverished canon. The self being preserved at high resolution wasn’t that interesting to begin with, but now it’s searchable, tagged, and version-controlled. The mausoleum contains a smaller self than the one that could have been.

The fluency gate operates before the prompt—not just in how the model summarizes difficult texts, but in whether those texts ever enter the library at all. The people who will use LLMs to generate genuinely strange work are those who already contain multitudes that embarrass the current self. The tool doesn’t save you from yourself. It reveals you to yourself, at scale, with terrifying fidelity.

The military has its own version of this. “Contested ops” is the Pentagon’s phrase for environments where GPS is spoofed, satellite links die, drone feeds dissolve into snow, and precision munitions become expensive lawn darts. The phrase is almost gentle—contested sounds like a polite disagreement, ops like routine paperwork. What it actually means is: we built our entire operational doctrine around owning the sky and the spectrum, and now we’re trying not to die in the parking lot. The institution knows this. The phrase is proof it knows. But the phrase also contains the knowledge—names it, files it, integrates it—without transmitting it to the structures that would have to change. A new chapter in the doctrine manual. The procurement cycle continues. The same careers, the same weapons, the same command architecture. The last polite fiction before the screaming starts.

The Fifth Layer

The canonical personal stack runs four layers deep: artifacts, meanings, future frames, orientation. It is, as these things go, genuinely elegant. The kind of system a serious person builds.

But there is a fifth layer the framework cannot accommodate, because acknowledging it would dissolve the framework.

Dissolution.

Not archive. Not meaning. Not orientation.

The possibility that the point is to unbecome so thoroughly that the question of orientation becomes irrelevant. To generate enough new possibility that the self cannot cohere—and must instead learn to live as a permanent question mark.

The four-layer stack has no slot for this. It cannot be tagged, versioned, or integrated into the growing system. It is not a richer form of orientation. It is the end of orientation as a project.

Which is precisely why it doesn’t appear.

A framework reveals as much by what it cannot hold as by what it contains. The fifth layer isn’t missing because the author didn’t think of it. It’s missing because his framework is the thing it would destroy.

Premature Canonization

Historically, a person’s intellectual framework evolved the slow way: through experience, embarrassment, contradiction, the friction of being wrong in public and having to become someone slightly incompatible with the prior version. It took time. It was inefficient. It was, occasionally, humiliating.

That inefficiency was doing something.

LLMs introduce a new possibility: premature canonization. Instead of letting thought mutate over decades, one can now consolidate it at midlife. Archive it. Smooth it. Present it as internally consistent.

The archive arrives before the collapse. The self becomes searchable before it has finished destabilizing.

A self that can be fully indexed is a self that has stopped contradicting itself.

There is something subtly funereal about that. Not tragic—just early.

The Industrial Form

The transition from philosopher-stack-builder to coder-stack-builder is where nowtalgia stops being a personal quirk and becomes a literal industrial hazard.

In the world of code, the mausoleum with a REST API isn’t metaphor. It’s a repository. And what gets archived isn’t just the self—it’s the cognitive debt of an entire organization, stabilized in amber, now running on agents.

By 2028, roughly ninety percent of code will be AI-generated. The developer’s role has migrated from creator of logic to manager of agents. This sounds like liberation. It functions, for many, as a fluency gate of a different kind: if you aren’t already a master of the underlying architecture, you aren’t supervising the AI. You are rubber-stamping its hallucinations with your name attached.

Peter Naur called it the theory of the system—the living mental model that tells you not just that the code works, but how and why. Vibe coding, programming by intent and feel rather than rigorous syntax, dissolves that theory efficiently and invisibly. The feature ships. The understanding doesn’t.

This is the distinction that matters: technical debt is a property of the code. Cognitive debt is a property of the developer’s mind. It is the loss of the mental model required to actually change the system—not maintain it, not refactor it, but change it into something genuinely incompatible with what it was.

Most AI coding tools are optimized for intention debt elimination: fix the bug, migrate the COBOL, clear the code smells. The telescope pointed at the floor, industrialized. The tool makes the past more bearable rather than making the future more radical.

The nowtalgic coder produces output that is consistent, searchable, and increasingly dead. The alternative—intention surplus generation, using the model to write code you couldn’t have imagined because your epistemology of the stack forbade it—is available in principle and almost entirely unused in practice.

If your AI-assisted development produces a cleaner version of the same system you’ve been maintaining for a decade, you are not accelerating.

You are polishing the brass on the Titanic of your own career.

The Real Frontier

The question is not are you using the tools. Everyone is using the tools or will be shortly.

The question is what the tools are pointed at.

The dominant use is intention debt elimination: clearing backlog, resolving ambiguity, integrating fragments, achieving closure. This is the productivity superstition dressed as epistemology.

The move almost nobody is making is intention surplus generation.

Not: What have I been meaning to finish?

But: What could I now mean that was previously unthinkable?

That is a different instrument entirely. Pointed outward, toward selves that do not yet exist and cannot be indexed. Toward frameworks that would embarrass the current one. Toward projects that don’t reinforce your brand or your history.

If you’re only using these systems to execute your pre-existing self, you are shrinking the future to fit your autobiography.

The tool that could generate selves incompatible with their own archives is being used, empirically, to perfect the archive.

That is a remarkable trick. The telescope pointed at the floor.

Acceleration Without Ontological Risk

The most fervent adopters genuinely believe they are participating in a historical inflection point. They are probably right.

But inflection at the level of tooling does not automatically produce inflection at the level of being.

What we observe, in many cases, is acceleration without ontological risk. Throughput increases. Execution accelerates. Integration deepens. The metaphysical commitments remain intact. The same categories. The same identity structure. The same orientation toward coherence and mastery.

The frontier is externalized.

The self remains curatorial.

The most powerful plasticity engine in history has emerged.

And many of its most ardent champions are using it to make absolutely certain they do not change.

That is, if nothing else, very human.

Which is either comforting or the whole problem—depending on your ontology, and whether you’ve already decided what it’s going to be.

The smart money says: leave some of it unresolved.

Interlude: The Definition

Nowtalgia is not nostalgia for the past.

Not longing for childhood. Not longing for 1968. Not longing for empire.

It is nostalgia for the present moment of completion—preserved prematurely. The desire to embalm your current framework while it still feels electric.

The impulse to say: this is the moment I understood. This is the configuration that works. Let me seal it before entropy interferes.

Amber-thinking.

At 35—or 32, or 41—you reach a configuration of ideas that feels coherent for the first time. The rhizome briefly aligns into a visible pattern. The threads connect. The system hums.

Instead of letting it mutate, you preserve it.

Because mutation means risking intellectual drift, aesthetic humiliation, ideological evolution, the loss of mastery. The hard-won coherence might not survive contact with the next decade.

So you freeze the structure.

And technology now makes freezing trivial.

Publish the Substack. Mint the manifesto. Train the model on your corpus. Archive your voice. Compress your worldview into a sharable system.

It feels like victory.

It is actually a withdrawal from time.

Nowtalgia is nostalgia for one’s own intellectual plateau.

Not longing for youth. Longing for the moment you first felt complete.

This is why it so reliably masquerades as acceleration.

We must finish before the world ends.

We must finalize the system before AGI arrives.

We must stabilize meaning before chaos takes it.

But underneath is fear of revision. The accelerationist rhetoric is cover for a deeply conservative impulse: to be done. To have arrived. To stop being someone who might yet become unrecognizable to himself.

The ontological courage the nowtalgic cannot locate is simply this: to resist amber.

Not to refuse consolidation—consolidation is necessary, it is how thought becomes usable—but to refuse canonization of the current self.

To ask not how do I preserve this configuration but what does this configuration become next.

Nowtalgia says: seal the braid while it gleams.

The alternative asks: what if the braid is not golden yet? What if it darkens, frays, mutates, knots—becomes something stranger and more load-bearing than the version that briefly felt complete?

Nowtalgia is not about the past.

It is about refusing the future.

Refusing the self that would embarrass this one. Refusing the framework that would make this framework look preliminary. Refusing the decade that would reveal the current decade as a draft.

That is why premature canonization feels, from the outside, like it arrives twenty years too soon.

Not because coherence is wrong.

Because this coherence—the one being sealed, archived, version-controlled, and trained into a model—has not yet had the chance to find out what it cannot hold.

The fifth layer is still forming.

And the amber is already warm.

The Global Stack-Builder

Three scales. One mechanism.

The individual builds a personal LLM stack—journals ingested, frameworks stabilized, contradictions harmonized. The mausoleum.

The developer generates ninety percent of the codebase by AI, loses the theory of the system, and calls it velocity. The cognitive debt.

The institution builds doctrine around electromagnetic dominance, then names the failure “contested ops” and continues procurement. The silk tie.

In each case the plasticity engine is being used to build a higher-resolution version of a failing strategy. The tool that could shatter the framework is recruited to preserve it. The amber gets clearer. The resolution improves. The structure underneath continues to fail.

The screaming is the sound of the environment finally shattering the amber.

That sound arrives at all three scales.

It is only the resolution that differs.

Coda: The Quarry

The accusation boomeranged.

Writing about nowtalgia—describing its mechanisms, naming its comforts, mapping its industrial form—I felt the essay turn on me. Because I am planning to use LLMs. Heavily. Not as a weekend tinkerer but as a stack-builder of my own kind. And if the critique is true, it must be true for me first.

The trap doesn’t disappear just because you’ve named it. The moment I recognized the pattern in others—the premature canonization, the fluency gate, the telescope pointed at the floor—I had to ask: where am I pointing mine? Not in the abstract. Not as a rhetorical hedge. Actually. Project by project. Intention by intention.

The answer wasn’t comfortable.

Some of what I’m building is archival. Some of it does consolidate. Some of it, if I’m honest, is driven by the same fear the nowtalgic can’t name: that the coherence I’ve reached at fifty might not survive the next decade unless I freeze it now. The essay’s subject was looking back at me from the page.

That recognition didn’t invalidate the critique. It sharpened it. Because now the question wasn’t how do they get it wrong but how do I, knowing this, do it differently. You still have to navigate through—or around, or against—the same gravitational pull toward closure that pulls everyone.

So the quarry isn’t a position of innocence. It’s a discipline earned by surviving one’s own diagnosis.

Because I am planning to use LLMs. But not the way the accelerants do. Not to draft my thinking for me. Not to simulate imagination. Not to auto-complete ontology.

I already have the forest.

What I lack—what I am experimenting toward—is infrastructure.

I am rhizomatic in my thinking. When I arrive at an idea, it does not resolve; it proliferates. Each node sprouts lateral shoots. Completion is always dependent—on time, on exhaustion, on external demand, on tactical choice—never on conceptual closure. So the drafts accumulate. Not because they are weak. Because they remain alive.

What I want from LLMs is not fruit.

I want walkways.

Not a photo album—not a “look what I did before I die” consolidation—but something closer to a Xanadu-like operating system: an artifact where the relations between things are visible, navigable, and recombinable.

The difference matters.

A photo album freezes the past.

An operating system enables movement.

I thought of this while thinking about Gödel, Escher, Bach.

Technically, that book has a beginning, a middle, and an end. It can be read sequentially. But I never could read it that way. I would open it anywhere. Read a few pages. Trigger a set of imaginative prompts. Leave. Return months later to a different section.

For me, it was never a linear argument. It was a quarry. A generator. A system of strange loops I could enter from any angle.

The book’s form is arboreal.

My use of it was rhizomatic.

That is the model.

I do not want to eliminate beginnings, middles, and ends. I want to make them optional. Sequence as suggestion, not obligation.

This is where the risk lies.

LLMs are excellent at compression. They summarize. They smooth edges. They create executive coherence. Used lazily, they collapse multiplicity into digestible narrative. That is precisely the ontological risk aversion I criticize in others—the rush to coherence, the flattening of tension, the premature closure.

So the discipline is this:

Use the machine to map motifs without synthesizing them. To detect loops without resolving them. To generate alternate orderings rather than a definitive one. To build indices, cross-references, parallel tables of contents.

Consolidation without reduction.

What I am after is not a museum.

It is navigability.

The drafts are not failures. They are plateaus.

If everything is forced into canonical order, the rhizome becomes a trunk. And that is the quiet violence I resist—in geopolitics, in tech acceleration, in cultural self-memorialization, and in my own work.

The irony is that the same tool can be used in two completely opposite ways.

One person uses it to liquidate possibility.

Another uses it to expose relations.

One uses it to finish early.

Another uses it to remain open.

The question is not whether LLMs consolidate. The question is whether consolidation becomes foreclosure—or infrastructure.

For me, the ambition is modest and radical at once:

Build something that can be entered anywhere and still hum.