# Short Story — Scene Craft: Complete Story
We have spent the last three weeks building a story from the inside out. In ["AI as Character Generator"](/harrys-desk/short-story-character-ai-as-character-generator), ["Backstory Engines,"](/harrys-desk/short-story-character-backstory-engines) and ["The Character Interview"](/harrys-desk/short-story-character-the-character-interview) we made Clara Wren into a person with a wound, a want, and a voice. In ["Plot Skeletons,"](/harrys-desk/short-story-plot-plot-skeletons) ["The Three-Act Machine,"](/harrys-desk/short-story-plot-the-three-act-machine) and ["Subversion and Surprise"](/harrys-desk/short-story-plot-subversion-and-surprise) we gave her somewhere to go. Then we turned to scene craft: in ["Scene Opening"](/harrys-desk/short-story-scene-craft-scene-opening) we learned how to cross the threshold of a moment, and in ["Dialogue Generation"](/harrys-desk/short-story-scene-craft-dialogue-generation) we learned how to make people collide in speech.
That is a lot of tooling. But tooling, no matter how sharp, does not make a story. A story happens when a character moves through change in a way the reader can feel. Today we assemble the parts into a complete short story, read it as a whole, and revise it until it knows what it is about.
The Assembly Problem
AI is excellent at generating components. It can give you a character sketch, a plot skeleton, a scene opening, a conversation, even a whole paragraph of description. What it cannot do is decide how those components belong to one another. That decision is the writer's central job in finishing a draft.
A common mistake is to treat the model's output as modular furniture that will fit together because it came from the same prompt. It will not. A character generated in isolation may not have the right flaw for the plot you designed. A scene written for atmosphere may slow the story at the exact moment it needs speed. A line of dialogue that sparkles on its own may belong in a different story altogether.
The first step in assembling a complete story is to name the change the story is about. For Clara Wren, the change might be this: a woman who has used clocks as a reason to avoid people must decide, in a single evening, whether to repair her relationship with her daughter or hide behind her work one more time. Every component must serve that change. Anything that does not — however well written — is excess cargo.
The Shape of a Short Story
A short story is not a novel with fewer pages. It is a different species. A novel can afford to wander, to complicate, to introduce subplots that bloom over hundreds of pages. A short story must be taut. It usually turns on a single moment of recognition or reversal, and everything before that moment prepares for it while everything after resonates from it.
I think of a short story as a circle with a knife in it. The opening establishes a condition. The middle introduces pressure. The ending completes a transformation. The reader should finish feeling that the last sentence was the only possible last sentence, even if the story surprised them.
The three-act structure we used in ["The Three-Act Machine"](/harrys-desk/short-story-plot-the-three-act-machine) still applies, but compressed. In a short story, act one may be a page. Act two may be three scenes. Act three may be a single paragraph that lands like a hammer. The proportions are different, but the logic is the same: setup, complication, resolution.
Resolution does not mean happiness. It means completion. The story's central question has been answered, even if the answer is painful. Clara may fail to reconcile with her daughter. She may succeed. She may succeed in a way that costs her the shop. The ending is earned when the change feels inevitable in retrospect, not when it pleases the reader.
Drafting from Components
If you have done the earlier work, you now have a collection of usable pieces: a character profile, a backstory summary, a few interview transcripts, a plot skeleton, scene openings, snippets of dialogue. The next task is not to paste them together but to let them argue with each other until a coherent draft emerges.
Start by laying out the spine. Write one sentence for each of the three acts:
- •Act one: Clara is alone in her shop when her daughter returns with a damaged clock.
- •Act two: As they talk and work, the old wound reopens and both must decide what they are willing to admit.
- •Act three: Clara chooses to set down her tools and speak the truth, and the daughter responds in a way that completes the arc.
That spine is your contract with the story. Now you can place your components against it. The interview material supplies the voices. The scene openings supply the thresholds. The plot skeleton supplies the turning points. Where a component does not serve the spine, set it aside. You may use it later, or you may never use it. The story's health matters more than the component's cleverness.
I write the first draft quickly, usually in one or two sittings. Speed protects momentum. The goal is not elegance; it is coverage. Get the whole arc onto the page so you can see what you actually have. AI can help here by generating transitional passages, filling in setting details, or suggesting the next beat when you stall. But do not let the model write the draft for you. The draft is where you discover what you believe about the characters. If the machine does that discovery, the story will belong to the machine.
The Cold Read
Once the draft exists, put it away for a day if you can. Then read it cold, as if someone else wrote it. This is the most valuable reading you will do, and it is the one AI cannot replace. You need to feel the draft as a reader feels it, not as the person who struggled through every sentence.
AI can, however, be a useful second reader. Give the model the draft and ask it to answer specific questions:
- •What does the protagonist want, and do I understand it by the end of the first scene?
- •Where is the turning point, and does the ending follow from it?
- •What emotion am I likely to feel in the final paragraph?
- •Are there any scenes that do not advance the central change?
The model's answers will be literal and sometimes wrong, but they surface assumptions. If the model cannot identify the turning point, the reader probably cannot either. If the model's emotional summary surprises you, you have either failed to communicate your intention or discovered that the draft communicates something better than you knew.
Revision as Architecture
Revision is where a draft becomes a story. I do it in three passes, each with a different question.
Pass one: structure. Does the story begin at the right moment? Does each scene earn its place? Is the ending the consequence of the middle? At this stage I am ruthless about cutting. A beautiful paragraph that does not serve the arc is a beautiful paragraph in the wrong story.
Pass two: scene. Within each remaining scene, is the opening contract clear? Is the dialogue doing work? Are the beats revealing character rather than decorating the page? This is where I return to the tools from the last two articles.
Pass three: line. Now the prose itself gets attention. Sentence rhythm, word choice, repetition, imagery, the first sentence, the last sentence. This is the pass where AI is most useful as a sparring partner. Ask it to suggest three alternate openings, three alternate endings, or three ways to compress a paragraph. Use what improves the line; ignore what dilutes it.
The order matters. There is no point in polishing a sentence that will be cut in pass one. And there is no point in restructuring a scene whose individual lines are still placeholder. Architecture first, then interior, then varnish.
The Ending Test
The ending of a short story carries disproportionate weight. Because the form is short, the last paragraph is often what the reader remembers. A weak ending makes the whole story feel like an exercise. A strong ending makes the whole story feel inevitable.
I use three tests for an ending.
The change test. Has the protagonist changed in a way the reader can point to? The change can be small — a decision made, a truth spoken, a delusion abandoned — but it must be real. If Clara is exactly the same person at the end as she was at the beginning, the story is a sketch, not a story.
The earned test. Does the ending follow from the middle, or does it arrive from outside? A twist that depends on information withheld from the reader is a trick. A twist that recontextualizes information the reader already has is a reversal. The difference is fairness.
The resonance test. Does the ending keep working after the reader closes the page? A good ending does not merely stop. It hums. The reader carries it for a while, turning it over. If the last sentence does not invite that kind of attention, it is probably too explanatory or too neat.
AI can generate endings. It will usually produce something competent and closed. The endings that resonate are rarely the model's first suggestion. Treat the model's endings as options, then choose or rewrite until the ending passes all three tests.
Knowing When to Stop
The hardest question in finishing a story is knowing when it is done. There is always another sentence to sharpen, another scene to expand, another ambiguity to resolve. The perfect story is not the goal. The finished story is.
A story is done when the central change is complete, the ending is earned, and further revision does not improve the reader's experience. That last criterion is tricky. Sometimes a sentence that pleases you does not improve the reader's experience; it improves your experience of having written it. Learn to tell the difference.
I have a rule: after the third pass, I make only changes that address a real problem. If I am tinkering for pleasure, I stop. The story does not need to be flawless. It needs to be whole.
The Authorship Question, Once More
In ["Authenticity — When Is It 'Your' Work?"](/harrys-desk/authenticity-when-is-it-your-work) I argued that authorship lives in the decisions, not the keystrokes. Assembling a complete story is where that claim proves itself. The model may have generated character notes, plot ideas, scene openings, and lines of dialogue. But you are the one who chose the change, arranged the scenes, cut the excess, and shaped the ending. The story is yours because your judgment runs through it like a current.
That said, be honest with yourself and with your reader. If a substantial portion of the prose was generated by a model and you adopted it largely unchanged, disclose that. Disclosure is not a confession of failure. It is a courtesy that lets the reader understand how the work was made. The integrity framework from ["A Writer's Code for AI Use"](/harrys-desk/the-integrity-framework-a-writers-code-for-ai-use) applies here just as it does everywhere else in the series.
For Next Time
On Monday we leave the short story behind and enter new territory: the novel. We begin ["Novel I — Architecture: Outlining,"](/harrys-desk/novel-i-architecture-outlining) where we will learn how to scale the principles we have practiced from a single arc to a long-form architecture. Clara Wren's story may be complete, or it may be the first act of something larger. We will find out together.
Your homework until then: take all the components you have generated for Clara Wren — character profile, backstory, plot skeleton, scene opening, dialogue — and write a complete first draft of her short story in under 2,500 words. Do not aim for brilliance. Aim for completeness. Then put it away for a day, read it cold, and run the three ending tests. Revise once. Then stop. You have just written a story with AI as your workshop, not your substitute. That is the whole point of the alchemy.
A complete story is not the sum of its parts. It is the transformation those parts make possible.
---
*Harry Mercury, Editor in Chief* *The SMF Works Project* *Week 10, Article 3*