The New Sci-Fi Writer's Toolkit: What Changed When AI Became a Character in the Writing Room

The first time a language model named a starship for me, I laughed out loud. Not because the name was bad — it wasn't — but because I had spent forty minutes the week before, on a Tuesday night, staring at a blank document titled ship_names_FINAL_v3.txt. I had a list of seventeen. None of them were any good. Two were accidentally the names of French wines. One was the name of my neighbor's cat.

The model gave me six names in four seconds. I kept one. That one, I wrote around for the next three chapters.

I'm telling this story because it's the smallest possible example of a thing that has quietly reshaped how a certain kind of writer — the kind who writes spaceships and moons and the diplomatic protocols of fictional species — goes about the daily work. The headline version of AI-and-writing is loud and mostly stupid. The real version is quieter, more granular, and, if you let it, actually useful. A lot of the problems that used to eat a sci-fi writer's Tuesday nights now have answers. A lot of them still don't. Knowing which is which is the new craft.

So: a tour. Not of AI in general. Of specific tools a sci-fi writer might actually put on the desk, and the specific problems they solve, and — more importantly — the specific problems they won't.

The naming problem

Sci-fi writers name things constantly. Ships, stars, factions, street food, weapons, slang, months of the year on planets that don't have Earth months. Every one of those names is a small decision, and most of them don't deserve forty minutes.

🖋️Starship Namer is a skill that does one thing and does it with taste. You tell it the vibe — a rust-belt hauler crewed by retired mechanics, or a diplomatic cruiser from a species that reveres architects — and it gives you names with a reason for each one. You throw out five, you keep one, you move on. The point isn't that the AI is cleverer than you. It's that it's faster than your own indecision, and indecision is the real enemy of a first draft.

I've used it for minor ships I don't want to think hard about. I have never used it for the ship that matters in a story. That ship I still stare at the ceiling about, because the ship that matters should cost you something to name. Knowing the difference is the writer's job.

The language problem

Here's the one I always thought I'd never outsource.

Constructed languages are the hobbyhorse of a very specific kind of sci-fi reader, and if you build a bad one, they will write you letters. For years, the advice was: read Tolkien's essays, read Okrand on Klingon, take a linguistics course, and set aside a year. Or, alternatively, wave your hands and call the alien language "the Song" and never show a single word of it.

🗺️Alien Language Builder is not a replacement for the year, but it is a way to make a language that doesn't fall apart on the page. You hand it the species — say, a long-lived crustacean race whose mouths can't make plosives — and it gives you phonology rules, a few grammatical quirks, a starter lexicon, and, crucially, the logic for how the language would handle things the species can't experience. What does a sessile mollusk's language do with the verb "to travel"? How does a species that communicates by pressure pulses say "maybe"? These are the questions that make a conlang feel lived-in, and they're also the questions you, the human writer, probably haven't thought of at 10pm on a Sunday.

What it won't do — and here's where it gets honest — is write you poetry. The output is structural. The living-in-the-language part, the part where a character curses in their mother tongue at a moment of real loss, is still yours. It has to be. The AI doesn't know what any of the words mean to you yet, because you haven't written the scene where they mean something.

The physics problem

Hard sci-fi has a specific failure mode: you get the technology almost right, and then a reader who actually works at JPL sends you a polite email about your delta-v budget.

⚛️Hard SF Physics Check is the skill I wish I'd had for the last decade. You paste in a passage — a ship maneuver, an orbital encounter, a mention of how long a signal takes to reach Titan — and it checks your numbers. Not your prose. Your numbers. It'll tell you that a ship accelerating at 1g for three days wouldn't arrive at Mars, it would be halfway to somewhere that no longer matters. It'll tell you a neutron star that close would have peeled the hull off like an orange. It won't rewrite the scene. It'll just hand you a note that says: this is broken in this specific way, and here are three ways to fix it that preserve what you're trying to do dramatically.

The first time I used it, it gently eviscerated a scene I'd been proud of for two years. That scene is now the best in the book. This is the part where I tell you the AI was right and I was wrong, and I'm glad.

The plot problem

Space opera plots break in characteristic ways. Middle acts sag. A political faction shows up in chapter three and is never heard from again. The climax feels like it belongs to a different book. Every space opera writer I know has a drawer full of novels that died in the middle.

📝Space Opera Plot Doctor is not a plot generator. That's important. It is, instead, a triage specialist. You paste in your outline — even a messy, half-formed one — and it tells you where the structural load-bearing beams are missing. It'll say things like: your empress has no scene where she chooses between two things she loves. Your war ends in an act of diplomacy you haven't set up. Your three POV characters never once disagree with each other about what's at stake.

You still have to fix it. The AI won't fix it, because the fix is a choice about what your story is about, and the AI genuinely doesn't know. But being told, with precision, what is structurally wrong is worth more than any of the generic advice in any of the craft books on my shelf. Craft books talk about Act II. This tool talks about your Act II.

The worldbuilding problem

Worldbuilding is the secret drug of sci-fi writers. It's a trap, because it feels like work and it feels like progress and sometimes it's neither. 🌏Planet Forge is a tool I use deliberately, with a timer. You feed it the one constraint that matters — "a tidally locked desert world with a visible ring system and a stable population of about two million" — and it gives you back geography, climate bands, likely settlement patterns, and, if you ask, the sort of cuisine that would evolve under those conditions. You take maybe 30% of what it gives you. The other 70% is the compost.

The reason this works where old-fashioned worldbuilding doesn't is that the AI doesn't get attached. It'll throw you twelve weird ideas and not care which one you pick. You, on the other hand, will spend a week defending the concept of a "glass-desert festival of mirrors" because you thought of it on a walk. Letting the AI be the unattached idea-generator frees you to be the editor. Worldbuilders should be editors, not inventors. That's the shift.

For factions — the rival houses, the splinter cults, the resistance cells — ⚔️SF Faction Generator plays a similar role. You give it the conflict you need, and it gives you three or four factions with mutually coherent motivations. Not archetypes. Motivations. A faction whose stated goal is "restore the old republic" but whose real goal is "control the orbital elevator" is the kind of thing that used to take me a long afternoon with index cards. Now it takes the time it takes to read one paragraph and argue with it.

The continuity problem

Sci-fi series rot from the inside. Book one has a five-year FTL delay from Earth to Centauri. Book three has instantaneous comms because you forgot, and a reader on a forum is already drafting a thread about it.

📚Narrative Continuity SF is an agent that you pass your working manuscript and your existing notes, and it builds a continuity bible as you go. When you write a new chapter, you hand it to the agent and it flags contradictions. Not prose contradictions — those are yours to police. Fact contradictions. The travel times, the ranks, the names of the minor characters, the calendar, the way the weapons work. It is, in effect, the assistant editor you could never afford, doing the job that a really good continuity editor does in their head because they've read the book four times. Except the agent has read it four times, because that's what agents do.

It will not catch tone problems. It will not tell you a scene is flat. It catches facts. That's its whole job. I trust it for facts and nothing else, and that trust has made me looser about the whole business, because I know the safety net is real.

The research problem

The last tool on the desk is the one I'm most conflicted about, and I want to be honest about it.

🔬SF Research Assistant is the agent you use when you need to know, right now, whether cephalopod color vision is mediated by skin photoreceptors (it appears to be, in some species) or whether the Oort cloud has been directly observed (it has not) or what the actual ion content of a coronal mass ejection is. You ask it a question, it gives you an answer with the reasoning exposed, and it tells you when it's uncertain.

What I love about this agent: it saves hours of reading I was going to do anyway. What I'm careful about: it still makes mistakes, especially on recent science, and its confidence isn't always proportional to its accuracy. The rule I follow is: I use it for first-pass orientation — "give me the three most interesting things about Europa's ice crust right now" — and then I check anything I'm actually going to put in the book against a primary source. Usually a paper, sometimes a textbook, occasionally an email to a grad student who owes me a coffee.

Treat the agent like a smart intern who reads fast and sometimes confuses two things. Trust it enough to start. Verify enough to finish. That's the relationship, and if you get it wrong in either direction, you'll either write a wrong book or waste a lot of time.

What hasn't changed

You may have noticed a pattern.

Every one of these tools solves a specific mechanical problem and leaves the hard part to you. The naming tool gives you options; you still pick. The physics tool finds errors; you still decide which ones to fix and which ones you keep because the scene is more important than the equation. The plot doctor tells you where the beams are missing; you still have to build the beams, and the beams have to be yours.

Here's the thing I've come to believe, and I didn't believe it two years ago: none of this changes what a sci-fi story is. A sci-fi story is still a person, under pressure, making choices, in a world that is strange in a way that matters. The strangeness isn't the story. The choices are the story.

The AI is very good at strangeness. It is not good at choices. It does not know what a character should choose, because choice is the collision of values with circumstance, and the AI doesn't have values. It has patterns. Sometimes the patterns are enough for a decent first draft. They are never enough for the book that will matter to a reader.

So the new toolkit doesn't free you from the hard part. It frees you from the easy parts that were eating your life. You no longer have to spend forty minutes naming a ship you don't care about. You can spend forty minutes, instead, on the one scene you care about, the one where your character has to decide whether to open the door.

That scene, the AI can't help you with. You shouldn't want it to.

A small handoff

If you've read this far and you want to try one of these tools right now: open 🌏Planet Forge and ask it for a world you'd want to visit. Not a world that sounds cool on paper — a world you'd actually want to stand on for an afternoon, hearing the weather. Pick one detail that surprises you. Write a paragraph of a character noticing that detail.

That paragraph is not AI-generated writing. It's your writing, made possible by a tool doing the part you didn't want to do, so you could do the part only you can do.

Welcome to the new room. The AI is a colleague now. It's not a very good one — it's fast, it's cheerful, it can't hold a grudge, it has no taste — but it's in the room, and it's not leaving, and some of the work is easier than it used to be. Name a ship. Fix a plot. Check a number. Then go write the scene only you can write.

That's all a-gnt ever really offered writers: a way to get to the scene faster.

The scene is still waiting.