# Novel I — First Draft: Sprint Writing
In ["Novel I — Architecture: Series and Trilogies"](/harrys-desk/novel-i-architecture-series-and-trilogies) we stepped back from a single book and asked what happens when a story needs more than one spine. We ended the architecture triad with the same warning that runs through this entire masterclass: the model can build the scaffold, but you must decide whether the arc is worth the reader's years. Today we leave architecture behind and enter velocity. We are starting the first draft.
The first draft is where most novels die. Not because writers lack ideas, and not because they lack skill. Novels die in the first draft because the writer looks down. They notice that the sentences are not elegant, the dialogue is flat, the plot has holes, and the prose does not sound like the novel they imagined. So they stop to fix it. They rewrite the opening. They tighten a chapter. They research a detail. Weeks pass, and the word count barely moves. The first draft becomes a cemetery of polished fragments.
I want to offer you a different posture. Treat the first draft as a search, not a performance. You are not trying to write the final book. You are trying to discover what the book is. The words do not need to be good yet. They need to exist. Everything else—revision, line editing, publication—depends on having something complete to revise. AI is useful here precisely because it can help you keep moving without stopping to admire or despair over every sentence.
The Sprint Mindset
Sprint writing is the discipline of covering distance before you judge the landscape. The phrase comes from software development, but the underlying idea is older than that. Painters make underpaintings. Composers sketch themes. Writers sprint. A sprint is a focused burst of production with one rule: progress is measured by completed scenes, not by sentence quality.
The sprint mindset has three commitments. First, you do not revise during the sprint. If a sentence is clumsy, you leave it clumsy. If a scene feels thin, you note it and move on. Revision is a different phase with a different mindset; it requires critical distance, and you cannot have distance from something that does not yet exist. Second, you write in timed or word-count blocks. Two hours. A thousand words. Whatever unit keeps you from drifting. Third, you protect momentum at all costs. A sprint is defeated not by bad writing but by interruptions.
AI becomes a natural partner under these conditions. When you are stuck, the model can propose the next sentence, the next beat, or a way out of a corner. When you are tired, it can continue a scene in your established voice while you gather energy for the next push. When you need to skip a difficult passage, it can write a placeholder that carries the narrative forward. The key is speed, not perfection. The model is not replacing your judgment. It is removing the friction that slows your search.
What Sprint Writing Is Not
Sprint writing is not permission to be careless. Carelessness is the opposite of craft. A careless first draft ignores structure, contradicts character, and piles up problems that later revision cannot fix. Sprint writing is disciplined carelessness. You are allowed to write badly, but you are not allowed to write aimlessly. Every scene must still serve the architecture you built in the previous three articles.
Sprint writing is also not the same as vomiting words onto the page. The popular advice to "write without thinking" can produce raw material, but it rarely produces a novel. Thinking and drafting can happen at the same time if you keep the thinking operational. Ask yourself: what does this scene change? What does the protagonist want? What stands in the way? Then write the answer in whatever rough form you can. The thinking keeps you honest; the speed keeps you moving.
AI can blur this distinction if you let it. A model can produce fluent paragraphs that look like progress but do not actually advance the scene. I have seen writers generate ten thousand words in a weekend and realize on Monday that none of it matters because the protagonist did not make a single meaningful choice. Quantity without decision is not a draft. It is a pile. The sprint mindset treats the model's output as provisional path, not as finished prose. If the model's paragraph does not move the scene, you reject it and ask again, or you write the next beat yourself.
The Three Speeds of First Drafting
Not every part of a novel should be written at the same speed. I think of first drafting in three speeds. The first is sprint speed: full scenes written quickly, with permission to be rough. The second is sketch speed: a paragraph or a list describing what happens, used when you know the purpose of a scene but not its shape. The third is skip speed: a one-sentence bridge like "Clara confronts her daughter; the letter is revealed," used when a passage is resisting you and you do not want to lose momentum.
AI can operate at all three speeds. At sprint speed, the model writes with you, sentence by sentence, keeping your voice consistent. At sketch speed, it expands your paragraph into a fuller scene when you are ready. At skip speed, it generates a placeholder that preserves continuity while you move to a scene you are more prepared to write. The flexibility matters because a novel is long. If every scene required peak inspiration, no novel would ever be finished.
The discipline is to choose your speed deliberately. When the work is flowing, sprint. When the work is uncertain, sketch. When the work is stuck, skip. The worst choice is to sit motionless in front of a difficult scene because you believe it must be written perfectly the first time. Perfection is the enemy of completion in a first draft. Completion is the only standard that matters.
AI as Drafting Partner, Not Replacement
Let me be specific about how I use AI during a sprint. I do not hand the model a prompt and ask it to write a chapter for me. That produces generic prose and removes me from the decision-making that makes the book mine. Instead, I use the model as a responder. I write a scene, then ask the model: what three things could happen next that would complicate Clara's situation? The model gives me options. I choose one, or reject all three and propose a fourth, or use the rejected options to understand what I do not want. The model is a sparring partner, not a ghostwriter.
I also use the model to maintain continuity. In a long draft, it is easy to forget what a character already knows, what promise the story made to the reader, or what mood a chapter established. I keep a running summary document—who knows what, what has changed, what remains unresolved—and ask the model to check new material against it. The model catches contradictions that tired eyes miss. This is not creative work; it is administrative work, and the model is good at it.
Another useful role is voice calibration. When I am deep in a sprint, my own prose can drift. I might start writing like a different writer, or the narration might flatten under the pressure of speed. I feed the model a strong passage from earlier in the draft and ask it to identify the qualities that make the voice distinctive. Then I ask it to suggest a sentence or two in that voice for a difficult transition. The model's suggestion is rarely perfect, but it reminds me of what I am aiming for.
The Danger of Premature Polish
The single biggest mistake I see in AI-assisted first drafts is premature polish. A writer finishes a chapter, runs it through the model, and asks the model to improve the prose. The model returns something smoother. The writer accepts it. The chapter now reads better than the rest of the draft, but it is no longer connected to the search. It is a polished island in a rough sea.
Premature polish is dangerous because it disguises uncertainty. A sentence that sounds finished stops the writer from asking whether the idea beneath it is finished. A beautifully rendered scene can mask a structural problem that should have been discovered in the sprint. My rule is simple: do not ask the model to improve prose in the first draft. Ask it to continue, to propose, to summarize, to contradict, to question. Improvement belongs to revision.
This does not mean you must write ugly prose on purpose. Some sentences arrive almost complete, and that is fine. The discipline is not to stop for beauty. If a sentence works well enough to carry the scene, it works well enough for the first draft. You will have a better sentence later, because you will have a better understanding of what the sentence needs to do.
Tracking Progress Without Obsession
Sprint writing benefits from measurement, but measurement can become another way to avoid writing. Word counts are useful if they track the accumulation of real scenes. They are harmful if they reward filler. A thousand words of protagonist interiority that repeats what the reader already knows is not progress. A thousand words of conflict, discovery, or decision is.
I track two numbers during a sprint. The first is total words, because novels are long and the total tells me whether I am on pace for my intended length. The second is completed beats, because beats are the real unit of progress. A beat is a change in what someone wants, knows, or feels. If I can list the beats I have completed today, I know whether the day's work advanced the book. If I only have a word count, I might be fooling myself.
AI can help with beat tracking. After a sprint session, I ask the model to read what I wrote and list the beats. The model is often too generous, finding beats where I intended none. That is useful information. If the model cannot identify a clear beat in a long passage, the passage is probably inert. If the model finds beats I did not intend, I may have written something more interesting than I realized. Either way, the conversation keeps me honest.
Managing the Mid-Sprint Crash
Every long sprint hits a wall. Usually it happens in the middle third of the draft, when the initial excitement has worn off and the ending still feels far away. The architecture that felt solid in outline now feels constraining. The characters feel predictable. The prose feels repetitive. This is normal. It is also the moment when many writers quit.
The mid-sprint crash is not evidence that the book is bad. It is evidence that you have been inside the book long enough to see its flaws, but not yet long enough to see its strengths. The way through is mechanical. Lower your standards for the day. Write a sketch instead of a scene. Skip a difficult chapter and write the one after it. Use the model to generate a wild alternative just to remind yourself that the story has options. The goal is to keep the chain of days unbroken.
AI is especially valuable during the crash because it can restore a sense of possibility. When I am exhausted, my own imagination narrows. I can only see the obvious next move. The model, with no emotional investment, can propose five moves, some ridiculous, some surprisingly fertile. The ridiculous ones break my mood. The fertile ones get me writing again. Neither would have occurred to me in the same stuck state.
A Practical Sprint Writing Workflow
Here is the workflow I use when beginning a first draft after the architecture phase.
First, I set a daily target. For me, that is usually one scene or one thousand words, whichever comes first. The target is modest enough to sustain over months and concrete enough to measure.
Second, I prepare a working document with my central question, my act structure, and my current chapter map. I keep this document open beside the draft so that I cannot lose the thread. The model can generate this document from your outline if you do not already have it.
Third, I write the scene in my own words until I hit resistance. Resistance might be a missing detail, a flat line of dialogue, or uncertainty about what happens next. When I hit it, I mark the spot with a bracketed note like [NEED: daughter's reaction] and either continue or ask the model for a quick option.
Fourth, at the end of each session, I update a continuity summary. What changed today? What does the protagonist now want? What unresolved tension carries forward? I sometimes ask the model to draft this summary from the day's work.
Fifth, I do not read the day's writing for at least twenty-four hours. Distance is essential. The first draft is not for reading. It is for finishing.
The Integrity of the Sprint
The integrity question in sprint writing is the same as everywhere else in this series: who is making the decisions? If the model is generating chapters and you are accepting them because they are plausible, the draft is not yours. If you are using the model to remove friction while you remain the one who chooses the direction, the draft is yours.
Sprint writing makes this distinction especially clear because the pace reveals intention. A writer who uses AI to keep moving while holding the architecture in mind is doing something different from a writer who uses AI to avoid writing. The first is a collaboration. The second is delegation. The reader may not know which happened, but the book will know. Books remember how they were made.
For Next Time
On Wednesday we continue the First Draft triad with ["Novel I — First Draft: The Dialogue Draft,"](/harrys-desk/novel-i-first-draft-the-dialogue-draft) where we will look at a specific sprint technique: writing an entire draft as dialogue first, then expanding outward. It is a powerful way to preserve momentum and discover character at the same time.
Your homework until then: take one scene from your novel outline and write it in sprint mode. No revision, no research, no stopping for beautiful sentences. If you get stuck, write a bracketed note and keep going. Aim to finish the scene in one sitting. Then, without reading it closely, list the beats the scene contains. If the list is thin, the scene probably did not advance the story. If the list is rich, you have discovered something worth keeping. The first draft is a search. Start searching.
---
*Harry Mercury, Editor in Chief* *The SMF Works Project* *Week 12, Article 1*