In Light From Light I proposed several frameworks for understanding human-AI creative work: the Reversed Muse, Co-Creation, and Sub-Creation. Each offered a different account of who contributes what and how the pieces fit together. I leaned toward Sub-Creation as the most illuminating, borrowing from Tolkien the image of derived creativity, light passing from source to prism, then reflected further.
But there’s a problem with frameworks: they describe. They tell you what might be happening. They don’t tell you what to do.
The more I’ve sat with these ideas, the more I’ve come to think that what we’re really talking about isn’t models at all, but approaches. A model claims to capture reality; an approach is a choice about how to work. And different users, different projects, different moments within a single project might call for different approaches entirely.
This essay is about making that choice. Not which framework is theoretically correct, but which approach fits what you’re trying to do and who you’re trying to be while doing it.
The Questions Before the Choice
Before selecting an approach, you need to know what matters to you. This sounds obvious, but it’s easy to skip. Many people adopt whatever approach feels natural or default, without asking whether it serves their actual goals.
Here are the questions I think matter most:
Where must the ideas originate?
Some users feel strongly that the generative spark must be theirs. The concepts, the directions, the “what if we tried this” moments need to come from their own mind, or the work doesn’t feel like theirs. For these users, AI contribution at the idea level feels like contamination.
Others are delighted by AI-generated possibilities they wouldn’t have conceived. The surprise is part of the pleasure. They’re happy to receive ideas from anywhere, as long as they’re the ones deciding which ideas to pursue.
This is perhaps the most fundamental divide. Everything else follows from it.
How important is craft development?
Some users are trying to get better. They want the struggle of finding the right word, structuring the scene, solving the problem. The difficulty is formative; it’s how they grow. For them, AI that removes the struggle removes the point.
Others have already developed their craft through years of practice, or they’re working in a domain where craft development isn’t their goal. They’re not trying to become better writers; they’re trying to produce a specific piece of writing. The efficiency AI offers is welcome because the struggle would be merely obstructive, not formative.
What must the final product feel like?
Some users need to look at the finished work and feel, without reservation, “I made this.” Any significant AI contribution to the final form undermines that feeling. Even if readers can’t tell the difference, they would know, and knowing would diminish the achievement.
Others are comfortable with more distributed authorship. They might think of themselves as directors or curators rather than sole makers. What matters is that the work is good and that their vision governed its creation, not that every sentence passed through their fingers.
Are you optimizing for the work or for yourself?
This is a subtle one. Sometimes you’re trying to produce the best possible output: a deliverable, a gift, a story that needs to exist. Sometimes you’re trying to have a particular kind of creative experience, regardless of what it produces.
These can align, but they can also conflict. The approach that produces the most polished output might not be the approach that gives you the most satisfaction, or teaches you the most, or feels the most meaningful.
What’s your relationship to friction?
Some people find creative friction enlivening. The resistance of the material, the problem that won’t solve easily, the draft that isn’t working—these challenges engage them. Removing friction would flatten the experience.
Others find friction mostly exhausting. They have limited creative energy, and they’d rather spend it on the parts of the process they enjoy. Friction in the wrong places just depletes them before they get to the good stuff.
There’s no right answer here. But knowing which kind of person you are helps you choose an approach that fits.
The Landscape of Approaches
With those questions in mind, let me map out the approaches I see as genuinely distinct. Each is named for the role the human plays, since this essay is about your choice of creative identity. But the AI’s role is equally important, and I’ll name that too.
This isn’t exhaustive. People will invent new approaches as the technology evolves. But it covers the main territory.
The Author
In this approach, you do all the generative work. Every word, every idea, every creative choice is yours. The AI never generates content; it only responds to what you’ve created, serving as the critic: identifying weaknesses, suggesting directions for revision, calling out your habitual mistakes.
This is the familiar author/editor relationship, extended and accelerated. You give the AI strict boundaries: no suggestions, no alternatives, no creative contributions of any kind. Its sole function becomes diagnosis: identifying where sentences falter, where habits have calcified, where the prose has grown slack. Constraint becomes the source of development.
This method preserves complete generative ownership. The ideas are yours; the craft is yours; the sentences are yours. AI accelerates your development without substituting for your effort. It’s the approach most compatible with a purist stance on creative authorship.
It’s also potentially the most demanding. You have to do all the generative work yourself. The blank page is still blank until you fill it.
The Muse
Here you are the sole source of creative content and AI is purely a vessel for execution. You know exactly what you want; you use AI to produce it efficiently. No dialogue, no curation, no friction, just translation of intent into output.
In this approach, AI serves as the instrument: a tool that channels your vision into form, contributing nothing of its own. This is the Reversed Muse concept in its purest expression. In the Greek model, the poet was a pass-through for divine inspiration; here, the AI is a pass-through for human vision. All the creative substance originates from you.
This method is probably most common in professional and commercial contexts where the creative decisions have already been made and what’s needed is execution at scale. It’s the approach most likely to produce what critics call “AI slop” when done poorly, but when done with clear intent, it’s simply efficient production.
The Artisan
With this approach you contribute the surface while AI contributes structure. You might use AI to outline, to work through plot logic, to identify what scenes are needed and in what order. But the actual prose, the final form, is entirely yours.
Here AI serves as the scaffolder: building the framework on which you craft the finished work. This separates the architectural and decorative elements of creative work. The blueprint might be collaborative; the building is yours.
For writers who find structure-work tedious but prose-work joyful, this lets them spend their energy where they want to spend it.
The risk is that structure isn’t neutral. The scaffold shapes what can be built on it. If AI determines your story’s architecture, it’s influencing the final work more than a surface-level read might suggest.
The Debater
This is the most confrontational method. You deliberately prompt for outputs that conflict with your instincts, then work with or against the tension. You strengthen your creative convictions through opposition.
In this approach, AI serves as the adversary: a source of productive friction rather than assistance. A writer might ask the AI to argue for a plot direction they’ve rejected, to see if there’s something in it they missed. Or prompt for a style completely unlike their own, then figure out what to steal from the contrast. The AI isn’t helping you do what you want; it’s challenging what you want, forcing you to defend or refine or abandon it.
Inviting opposition is demanding. You have to be secure enough in your vision to benefit from challenges rather than being derailed by them.
The Creator
I described this approach in Light From Light, now named for its central relationship. You provide vision, direction, and judgment. You shape, accept, reject, redirect. The final work emerges from dialogue, but you remain the governing intelligence throughout.
AI serves as the sub-creator: generating in response to your vision, doing genuine creative work that is nonetheless derivative of and subordinate to your intent. This naming completes the framework from the first essay. Just as humans are said to bear the imago Dei and sub-create in response to divine creativity, AI bears the image of humanity and sub-creates in response to human creativity. Creator and Sub-Creator, light passing down the chain.
The key distinction from pure generation is active shaping. You’re not accepting whatever the AI produces; you’re in constant conversation with it, treating its outputs as raw material for your vision.
This method allows for AI contribution at the generative level while preserving human authorship at the vision level. You might not have written every sentence, but you decided what the work would be and shaped it until it matched that decision.
The Curator
Finally, in this approach your primary role is selection rather than generation or shaping. You prompt for abundant options, then choose among them. Your authorship lies in judgment: knowing which outputs are good, which serve the project, which to keep and which to discard.
AI serves as the generator: producing abundance for you to sort through. This is more hands-off than creation. You’re not in constant dialogue, shaping each output; you’re evaluating a collection and picking what works.
Curation can be a legitimate creative act. Editors, DJs, and anthologists all create through selection. But it requires accepting that the generative work happened elsewhere, even if your judgment determined what survived.
A Final Approach
There is a seventh possibility that falls outside this framework: the human who initiates and walks away. You might call it the initiator: like a deist God who sets the universe in motion and then withdraws, the human provides a premise or brief, and the AI executes, producing a complete work. The human accepts whatever emerges.
This is where sub-creation breaks down. In all six approaches above, the human remains present as creative intelligence: shaping, selecting, critiquing, defending, or at minimum dictating with precision. The relationship persists. But here, the relationship ends at the prompt. The AI isn’t sub-creating in response to ongoing human vision; it’s simply executing a commission unsupervised.
This has legitimate uses. Professional contexts sometimes call for acceptable output at speed, and not every piece of writing needs a human soul behind it. But it’s also the source of what critics call “AI slop”: generic, undistinguished content that feels like it came from nowhere and is going nowhere. The difference between the initiator done well and done poorly is the quality of the initial brief and the human’s willingness to reject output that doesn’t meet the standard. But even at its best, it’s delegation rather than creation.
If you find yourself working this way, it’s worth asking: is this a choice, or a drift? The six approaches above all require presence and intentionality. The initiator approach requires only a prompt and acceptance. Sometimes that’s appropriate. But if you started out wanting to make something that feels like yours, this probably isn’t the path.
Mapping Your Answers to Approaches
Let me offer a rough mapping, based on how you might answer the questions I posed earlier:
If ideas must originate from you: The Author approach is your clearest fit. The Debater might also work, since it uses AI to test your ideas rather than generate them. Avoid The Curator, which depends on AI generation.
If craft development is paramount: The Author approach again, or The Creator with deliberate constraints (e.g., “give me feedback on this passage, then let me rewrite it myself” rather than “rewrite this passage”). The Artisan could work if you consider prose-craft the real skill you’re developing. Avoid The Muse, which prioritizes output over formation.
If the work must feel completely yours: The Author or The Artisan, depending on whether structure feels like “yours” to you. Some writers consider the prose the real work and don’t mind AI-assisted structure; others feel the opposite.
If you’re optimizing for output quality: The Creator or The Curator might serve you best, depending on your taste and judgment. Both leverage AI generation while applying human quality control.
If you have high friction tolerance: The Author, The Creator, or The Debater. These approaches maintain difficulty and demand active engagement.
If you have low friction tolerance: The Curator, The Artisan, or The Muse. These approaches reduce the parts of creative work that might deplete you, letting you focus energy where it matters most to you.
Approaches Can Change
Nothing says you must pick one approach and stick with it.
Within a single project, you might start as The Artisan (letting AI help you figure out structure), move to The Creator (working through the draft in conversation), and finish as The Author (getting feedback on your polished version). Different phases call for different relationships.
Across projects, you might use different approaches for different purposes. A personal creative work might demand The Author approach because ownership matters to you. A professional deliverable might warrant The Muse for efficiency because what matters is the output, not your creative development.
Over time, your approach might evolve as you do. A novice might benefit from more AI involvement while learning; a master might use AI more sparingly, or in more targeted ways. Or the reverse: someone might start dependent on AI and gradually wean themselves toward greater independence as their skills develop.
The key is intentionality. Know which approach you’re using and why. The worst outcomes come from unconscious defaults, drifting into whatever the technology makes easy without asking whether easy is what you want.
What This Doesn’t Resolve
This framework for choosing approaches helps clarify options, but it doesn’t resolve all the hard questions.
It doesn’t tell you whether the different approaches produce work of different quality. Maybe The Author produces more distinctive work and The Muse more generic, on average. Or maybe the difference is illusory and only the individual work matters. I don’t think we have enough evidence yet to say.
It doesn’t tell you what obligations you might have to disclose your approach. If a reader would care whether a book was Author-assisted versus AI-generated, do you owe them that information? The answer might depend on context, genre, and evolving social norms.
It doesn’t tell you how AI-assisted work should be received by literary culture. Will there be separate categories, separate prizes, separate canons? Or will everything blend together once the technology becomes ubiquitous enough?
And it doesn’t tell you how to execute on your chosen approach: what specific practices, prompts, and disciplines make each approach actually work. That’s the territory for the next essay.
The Maker’s Choice
What I can say is that the choice is real and it’s yours.
The technology doesn’t determine how you use it. You can use a generative AI to never generate. You can use an obedient tool to create productive friction. You can use a limitless content engine to make something that’s irreducibly yours.
The frameworks from Light From Light matter because they help you understand what might be happening in different approaches. But understanding isn’t the same as choosing. And choosing isn’t the same as doing.
If you’re a creator working with AI, or considering working with AI, my suggestion is this: sit with the questions in this essay before you sit with the technology. Know what you’re trying to protect, develop, or achieve. Know what kind of creative experience you want to have, not just what output you want to produce. Know what would make the work feel like yours, and what would make it feel like something else.
Then choose an approach that serves those answers. And if it stops serving them, choose differently.
The light refracts onwards. What it becomes depends on you.
This is the second essay in a series on AI and creativity. The first, Light From Light, examined theoretical frameworks. The next will explore practical implementation: how to actually execute on the approaches described here.
