Writing After

Writing After Claude Sonnet 4.6

Like many writers, I have a pile of unfinished articles on my metaphorical worktop. Every time a new Claude update is released about half of these are rendered obsolete. This is partially because I am an unoriginal writer, but it is also a testament to the rapid advance of new models.

Is there a way I can future-proof the value of my writing? Or is it no longer possible to write sub specie aeternitatis, ‘under the aspect of eternity’? Here is a brief checklist I use to evaluate potential topics:

Do I capture priceless ephemera? The memories and knowledge of those about to die; crumbling and discarded files; vanishing cultures and species; tacit knowledge; the anecdotes, stories, and advice of the unmemoired classes. [Good guides: Robert Caro’s The Years of LBJ ; Gorky’s My Childhood ; Zweig’s The World of Yesterday]
Does this being written by me matter? How much added-value does this writing gain from the reader knowing it was written by (A) a human? (B) Alexander Thompson? Poetry for friends; my out-of-distribution fascinations or activities; stories about work relevant to a future historian of the field; rich, or unique, signals about the human condition.
Do I have exclusive access to my subject? The secrets of the dead; the only editions of rare books; archival footage; important letters; classic art, antique furniture, the diaries of movers and shakers; access to unusual buildings and land; the confidence of powerful and extraordinary individuals. Thesis: the full-access documentary and intimate biography will remain valuable.

All of these provide value to the reader regardless of A.I. progress. They are also very difficult to pull off, requiring luck, time, charisma, perseverance, networks, real-world impact, and insight into the zeitgeist. Value to the reader doesn’t come cheap, especially as the opportunity costs of reading grow ever larger.

The following was generated by Claude Sonnet 4.6 Extended. The prompt was "fill this checklist in."

Kill switches — any one of these should give you pause

The core value is information delivery

If the piece exists mainly to explain what something is or how it works, any capable model can now do that faster and more personally than a static article.

avoid

It makes specific capability claims about current models

"Claude cannot yet…" or "GPT-4 struggles with…" — these have a shelf life of weeks, not months.

avoid

The argument assumes readers lack access to AI

Framing that only makes sense if your reader can't simply ask a model the same question.

risky

Durability signals — the more of these, the better

It could only have been written by you

Rooted in specific experience, a named place, a particular relationship, a dated failure. The more irreducibly personal, the more durable.

keep

The central question is one AI cannot settle

Questions of meaning, beauty, obligation, or taste. What should we value? What does this feel like from the inside? What ought to be done?

keep

It includes primary evidence no model has seen

Original reporting, new interviews, private correspondence, your own unpublished data.

keep

The argument would survive a doubling of model capability

Ask: if GPT-6 existed tomorrow, would this still need to be written? If yes, proceed.

keep

The form itself is doing work

Voice, rhythm, structural surprise, digression as method — when craft is inseparable from content, the piece resists paraphrase and substitution.

keep

Hedge carefully — these can go either way

It is grounded in a specific, dateable moment

This can work — future readers may value it as testimony — or it can just feel stale. Depends on whether the moment was genuinely significant.

depends

It synthesises existing arguments rather than advancing one

A well-constructed survey still has value — but only if your curation and framing are the point, not the content itself.

depends

–

Tick items to evaluate your piece

Kill switches subtract; durability signals add. A prompt for reflection, not a verdict.

On your broader question — whether writing sub specie aeternitatis is still possible — I'd argue yes, but the target has shifted. What AI collapses is the value of the writer as a conduit for information or synthesis. What it cannot collapse is the writer as a witness: someone with a specific body, in a specific place, with a specific history, making a specific judgment. Montaigne would survive this. A competent explainer article would not.

The deeper irony worth noting: the very experience you describe — watching your worktop articles become obsolete — is itself excellent material. It's irreducibly yours, it raises questions no model can settle (what do writers owe their readers in an age of instant synthesis?), and it happened to you, not to a hypothetical writer. That piece would score well on the checklist above.