What changes for writing with AI?

The shift

Producing a competent first draft — in any register, length, or language, from a blank page or from source material — goes from scarce (a writer's trained hours) to abundant and near-instant. What stays scarce is knowing whether the draft is true, whether it's any good, and who stands behind it once it's published.

The axioms

• Writing well requires years of practice most people haven't put in. Rests on drafting competence being scarce and slow to acquire.
• A blank page is the hardest part; getting something down is most of the labor. Rests on first-draft generation being effortful and time-consuming.
• Editors and agents exist because most writing needs an outside eye before it's ready. Rests on second-pass judgment being expensive to get and in limited supply.
• A writer's voice — the specific way they use language — is a durable asset that took time to develop. Rests on distinctive style being rare and hard to imitate.
• Volume is capped: one writer produces a finite number of usable words per day. Rests on production capacity being bound to a single person's hours.
• Research and fact-gathering before writing take real time. Rests on synthesis of source material being slow and effortful.
• A byline signals a person is answerable for what the piece claims. Rests on accountability being tied to an identifiable, liable individual.
• Readers can generally trust that words on a page reflect someone's actual thought or effort. Rests on the cost of producing fluent text being high enough that its existence implied effort.
• Editing (as a craft distinct from drafting) is cheaper and faster than writing from scratch. Rests on the asymmetry between generation cost and revision cost.
• Publishing something is a costly enough act that most of what gets published cleared some bar. Rests on distribution and production together gating volume.

Invalid axioms

1. Writing well requires years of practice most people haven't put in. Producing fluent, structurally competent prose in any register is now free and instant — the skill that used to gate who could produce a readable draft is commoditized. Habit-trap: hiring, training programs, and status still treat "can write a clean paragraph" as the scarce credential, when that's now table stakes a model clears by default.
2. A blank page is the hardest part; getting something down is most of the labor. Draft generation is no longer the bottleneck in any writing process — a usable starting point exists in seconds. Habit-trap: workflows, timelines, and self-image ("I'm a slow starter") are still built around the drafting phase being the hard, time-consuming part, when it's the part that's now free.
3. Volume is capped: one writer produces a finite number of usable words per day. Output volume is no longer bound to a single person's hours — the same person can generate, in parallel, more draft material than they could physically type. Habit-trap: word-count-based pay, page-rate freelance pricing, and "how many pieces can you turn out a week" staffing models still price the scarce resource as words-per-hour when the actual constraint has moved elsewhere.
4. Research and fact-gathering before writing take real time. Synthesizing a pile of source material, prior coverage, or reference documents into a usable summary is now near-instant. Habit-trap: schedules and rates still budget a research phase as heavy, unstructured human time, when the synthesis step itself is now cheap — only the verification of what it produced is not.
5. Editing (as a craft distinct from drafting) is cheaper and faster than writing from scratch. With generation this cheap, "just regenerate a fresh draft against better instructions" often beats line-editing an existing one. Habit-trap: revision workflows still assume the sunk draft is worth preserving and polishing, rather than treating any given draft as disposable and re-generating from a sharper brief.

Unchanged axioms

1. A byline signals a person is answerable for what the piece claims. A model can produce a fluent, well-sourced-looking paragraph that is fabricated, misattributed, or subtly wrong, and it cannot be corrected against, sued, fired, or lose standing. Accountability for what a piece asserts still requires a named human who can be held to it — this doesn't get cheaper just because drafting did.
2. A writer's voice is a durable asset. Models can imitate a style on request, but choosing which voice serves a specific reader, occasion, and stake — and recognizing when a piece has drifted off-voice — is a judgment call, not a pattern-match. The imitation is abundant; the taste that picks the right voice for the moment isn't.
3. Editors and agents exist because most writing needs an outside eye before it's ready. The specific failure mode of a model — confidently fluent and wrong, or fluent and flat — is exactly what a second, differently-motivated human reader catches. That review function gets more necessary as draft volume increases, not less.
4. Readers can trust that words on a page reflect someone's actual thought or effort. This was never fully true, but the cost floor that made mass-produced fluent nonsense rare is gone. What's scarce now isn't the ability to write sincerely — it's a reliable signal that a given piece of writing did.
5. Publishing something is a costly enough act that most of what gets published cleared some bar. The bar was always uneven, but production cost was doing real filtering work. With that filter gone, the judgment call about whether something is worth publishing — not whether it can be produced — is the scarce gate now, and it still requires a human decision.

New axioms

1. Verifying fact-heavy or high-stakes writing at the volume AI-assisted drafting now enables. When ten drafts can be generated in the time one used to take, fact-checking and source-verification capacity doesn't scale with it — a bottleneck that didn't bind as hard when draft volume was naturally throttled by human typing speed.
2. Telling apart writing someone actually thought through from writing that's merely fluent. When fluency stops correlating with effort or original thought, readers, editors, and hiring managers lose a signal they used to rely on without noticing they relied on it — there's no settled replacement yet.
3. Pricing and structuring writing work when the drafting step is free. Page rates, per-word fees, and "billable hours of writing" assumed drafting was the costly part. What a client is actually paying for now — judgment, research verification, accountability, editorial taste — needs a different unit than words produced, and most rate cards haven't caught up.
4. Training new writers when the hardest visible part of the craft is now automated. Junior writers used to learn structure and voice by grinding through the effortful parts of drafting; if that grind is skipped, it's unclear what replaces it as the way someone develops judgment, taste, and a durable voice of their own.
5. Distinguishing a person's writing from a model's output at the scale platforms and inboxes now see it. Cover letters, reviews, comments, articles, and pitches arrive in AI-assisted volumes indistinguishable from effortful human work, and neither readers nor the platforms carrying them have a reliable way to tell — a provenance problem that barely existed when producing fluent text took real time.

Where it breaks

Writing programs and hiring pipelines are cutting the drafting-practice grind because it's no longer the visible bottleneck (INVALID: years-of-practice-to-draft-well was never actually the scarce resource once generation got cheap) at the same moment nobody has a working substitute for how a writer develops judgment and voice without that grind (NEW: the training-without-the-grind problem is unsolved). The stage that used to build the scarce skill is being removed to save time on a skill that just stopped being scarce.

Second collision: publications and platforms are leaning harder into "more content, faster" now that drafting is free (INVALID habit — still resourcing as if volume were the win, when volume was never the constraint that mattered) while readers and platforms have no reliable way to tell which of that content had a verified, accountable human behind it (NEW: the provenance problem). Producing more only compounds the trust deficit it's trying to outrun.

ai-axioms.vercel.app/axioms/media/writing-ai-axioms