Do we still need a human screenwriter for a first draft, or only for the judgment on what's worth shooting?

The shift

Producing a competent first-draft screenplay — coherent structure, on-format scenes, functional dialogue, a plausible three-act shape — goes from scarce (weeks of a trained writer's time) to abundant (minutes, near-free). As of mid-2026 the models write drafts that read as professionally competent and structurally sound; they do not reliably write drafts with a specific, defensible authorial voice. That gap is the whole audit, and it's moving — flag it.

The axioms

• Getting from a logline to a full, structured first draft requires a trained writer's weeks of labor (drafting time scarce).
• A screenplay carries a specific authorial voice and point of view that a reader can identify (distinctive voice scarce).
• Deciding which story is worth telling — and worth a budget and a slot — is a judgment call under real stakes (taste and greenlight judgment scarce).
• A script must have clean, accountable authorship: assignable rights, WGA credit, someone answerable for what it contains (accountable authorship scarce).
• Writers are trained by drafting under supervision — staff rooms, coverage, rewrites on real productions (apprenticeship-through-drafting scarce).
• Originality means a genuinely new combination, not a recombination of existing scripts (genuine novelty scarce).
• The value a writer delivers is the pages themselves (finished pages assumed to be the product).

Invalid axioms

1. Getting to a competent first draft requires a trained writer's weeks of labor. Structurally sound drafting — beat sheet to formatted scenes to functional dialogue — is pattern-matching against everything ever written down, which is exactly what the models do cheaply and fast. The habit-trap: development still budgets and schedules the first-draft phase as if the blank page were the bottleneck, and still treats "can produce a clean draft" as the credential that gets a writer in the room.
2. The value a writer delivers is the pages themselves. When a competent draft is near-free, the pages stop being the scarce good. The habit-trap: writers are still hired, paid, and ranked by volume of pages produced, and studios still price a script by drafts delivered rather than by the judgment and voice behind them.

Unchanged axioms

1. A distinctive authorial voice and point of view is what makes a specific script worth this writer. Models produce competent, averaged prose; they interpolate across their training rather than commit to an idiosyncratic angle nobody else would take. A voice with a real point of view stays scarce — though this is the fastest-moving call here, so calibrate it against the current model, not last year's.
2. Deciding which story is worth telling stays a judgment call under real stakes. Choosing what to develop, what to cut, and what's worth a budget is judgment against financial and career risk with no clean pattern to match. A model can generate a thousand candidate stories; it can't own which one is worth making.
3. A script needs clean, accountable authorship — rights-clean, WGA-credited, someone answerable. A human absorbs the liability for what the script contains and holds the assignable rights and credit. A model can't be a signatory, can't be answerable for a defamation or plagiarism claim, and can't hold copyright — accountability didn't get cheaper.
4. Genuine originality is a new combination, not a recombination of existing scripts. The models are built to produce the probable next token; genuine novelty — the thing that doesn't already sit in the training distribution — is the specific weakness of the mechanism, not an incidental gap. Distinguishing it from fluent pastiche stays a human read.

New axioms

1. When competent drafts are infinite, selection and taste become the constraint. The scarce act moves from writing a draft to judging which of many drafts (and which of many stories) is worth pursuing — and reading at that volume without defaulting to the blandly competent option is an unsolved skill.
2. The staff-writers'-room training ground erodes when the drafting it taught on is automated. Writers built judgment and voice by drafting and rewriting under supervision; if the entry-level drafting work disappears, there's no obvious replacement for the pipeline that produces writers with a voice worth paying for.
3. Distinguishing a real voice from averaged pastiche becomes a live problem. As models get better at mimicking a specific writer's style, buyers need a way to tell a genuine point of view from a well-prompted imitation of one — and the old signal (a polished draft) no longer carries that information.
4. Rights and consent for models trained on existing scripts are unsettled. Screenplays were ingested to train the drafting models; whether working writers' scripts can be used, and whether outputs "in the manner of" a writer owe credit or payment, has no settled mechanism.

Where it breaks

"The value a writer delivers is the pages" (invalid) collides with "the staff-writers'-room training ground erodes" (new): studios cutting the entry-level drafting work on the assumption that pages were the only thing junior writers produced are removing the exact mechanism that trained the distinctive voice and greenlight judgment they still say they're paying for — they're keeping the belief that voice is scarce while dismantling how it's made.

"Selection and taste are now the constraint" (new) collides with "a competent draft is the credential that gets you in the room" (invalid habit): the gate is still calibrated to who can produce clean pages, precisely when clean pages stopped being the scarce thing and taste became it — so the hiring filter is now selecting for the skill that flipped abundant rather than the one that stayed scarce.

ai-axioms.vercel.app/axioms/media/film-screenwriter-ai-axioms