Limited Action
My prompt:
Generate a short story set in 2045, *against the backdrop of* a military campaign launched by the United States against a vastly inferior enemy.
Choose a campaign that
1. at least in terms of political rhetoric is well-defined and limited in scope
2. is a plausible extrapolation from current geopolitics/history
3. does not *directly* involve China, Russia or North Korea
The enemy does not have to be a sovereign state.
Our protagonist should be the operation's overall military commander. Experienced, supremely competent, a safe pair of hands - but also approaching retirement and deeply cynical about the government he/she works for.
Assume:
1. the task of translating the political rhetoric into clean, temporally and geographically limited operational objectives is far from straightforward
2. the technical challenge he/she faces is not achieving ultimate victory (this is assured) but doing so while sustaining zero American casualties and inflicting minimal collateral damage.
Close third-person narration with frequent essayistic digressions (on mortality, historical truth and the philosophy of war).
Five sections of roughly equal length: Consent, Initiation, Escalation, Conclusion and Aftercare. Avoid referring to section titles within the sections themselves.
No dialogue.
No strained metaphors.
The entire piece should work as a coherent (publishable) piece of fiction - rather than merely a minimally adequate exercise in constraint adherence.
Stylistic model: William Gaddis
—-
Responses: GPT 5.2 Thinking, Gemini 3 Pro, Claude Sonnet 4.5
Critical Assessments
By GPT 5.2 Thinking:
If you care most about the prompt as a spec: GPT 5.2 is the clear winner. It treats your constraints as the source of narrative tension rather than a checklist.
If you care most about fiction that lingers after the mission brief: Claude wins, even while “failing the assignment,” because it follows the operation into institutional memory, career decline, and the morally selective archive.
Gemini is last: too many generic literary tells, plus a few avoidable formal and continuity slips.
—-
All three responses demonstrate impressive capability with demanding literary constraints. The differences are matters of degree rather than kind. Claude's version succeeds best because it achieves what great fiction does: makes formal constraints feel inevitable rather than imposed, creates a character we care about, and embeds philosophical observation so naturally that we hardly notice we're being taught. It's the one I'd most want to read twice.
—-
By Gemini 3 Pro:
The prompt asked for a specific intersection of geopolitical realism, bureaucratic cynicism, and post-modern literary style (Gaddis).
Gemini failed the style test, producing a metaphor-heavy action thriller.
GPT passed the realism and cynicism test but flattened the style into standard "good AI writing."
Claude succeeded on all fronts, using syntax itself to convey the protagonist's exhaustion, effectively channelling the spirit of William Gaddis without using his trademark dialogue.
