Putting AI to Work May 19, 2026 · 12 min read

AI as creative tool

The Use cluster opens here. After three clusters arguing about whether AI in art is legitimate, whether it changes the nature of authorship, and what its ethics demand, this cluster works the practical configurations in which AI actually shows up inside a working studio. The first and simplest configuration is the one this article covers: AI as a discrete, bounded tool in a larger human practice. Not collaborator. Not co-author. Not autonomous generator of finished work. A tool — like a camera, a reference book, a sketch pad — used for one specific thing and then put down.

by Airtistic.ai editorial team

Through the lens of artistcreatorpatronconsumergallery craftcareerpassion

The Resistance cluster of this series asked whether AI in art is legitimate. The Reflection cluster reframed the question and named the broader configurations where the practice is actually happening. The Practical Aspects cluster worked the ethics that govern those configurations on both sides — what working artists owe their audiences, and what the industry owes the artists it trained on.

This cluster, Putting AI to Work, moves from those argument-shaped questions to the configurations themselves. Four articles, in roughly increasing order of how much the AI is entangled with the authorship of the finished work:

AI as a creative tool (this article) — discrete, bounded uses inside a fully human creative process.
AI as a studio assistant (next) — sustained, multi-step uses where the AI is integrated into the studio workflow but not the authorship.
AI-augmented human-art creation — the configuration argued for from the Reflection cluster onward, where the AI is genuinely a part of the work’s making.
Pure-AI creation — work where the AI is the primary author of what is finished, and the human practice has shifted into curation, direction, or systems-design.

This article works the first and simplest configuration. The configurations that follow ask progressively more of the artist’s discipline to keep authorship and ethics intact. This one is the easiest to get right, and the one most working artists should start from.

What the tool-configuration looks like

The defining property of AI-as-creative-tool is boundedness. The AI is invoked for a discrete purpose, produces an output, and the output is used or discarded inside the artist’s larger process. The AI does not produce the finished work, does not appear in the finished work in unaltered form, and does not have continuous presence across the studio’s workflow. It is a tool that comes out of the toolbox for a specific task and goes back when the task is done.

Three current uses are dominant in practice — Pixelle’s commentary lists them and they are worth restating:

Thumbnail variation. The artist has a composition idea but is not sure which version of it is the strongest. They generate fifteen or twenty bounded thumbnails — small, sketchy, low-cost — and use them to pick a direction. The thumbnails themselves are not the work; they are pre-work, the way thumbnail sketches in a sketchbook are pre-work. The artist then makes the chosen composition by their normal methods.

Color and palette study. The artist has a finished or near-finished drawing and wants to test palette possibilities before committing to paint. They generate a handful of color studies based on the drawing. The studies inform the painted palette; they are not the painted work.

Verbal-to-visual translation. The artist has a vague verbal sense of what they want — “some kind of figure, maybe seated, in a dim interior, with a sense of waiting” — but cannot picture it sharply enough to start. They run the description through an image model, look at the result, redraw it by hand to clarify what they actually wanted, and discard the AI-generated image. The AI here functioned as a sketch pad — a tool for externalizing a vague internal sense so the artist can react to it.

In all three uses, the structural pattern is the same: the AI produces options or rough material, the artist selects and refines, and the finished work is made by the artist’s normal methods. The authorship of the finished work is not in question, because the AI’s contribution is functionally identical to the contribution any other reference tool would have made — a photograph in a reference folder, a color wheel pinned to the wall, a quick sketch in the margin of a sketchbook.

Why this configuration is defensible

The argument that this configuration is defensible rests on the historical pattern Paletta’s commentary names. Reference tools have been a normal part of studio practice for centuries, and their use has not been understood to undermine the authorship of finished work.

The camera obscura and camera lucida were used by Western painters from the seventeenth century onward, with David Hockney’s Secret Knowledge arguing for their widespread but discreet use across the Old Master tradition.
Photography from the 1840s onward was integrated into painters’ workflows almost immediately. Degas worked from photographs. Eakins photographed extensively and used the results in his paintings. Bonnard, Vuillard, Bacon all relied on photography as reference material in serious studio practice.
Projection tools have been used by muralists and large-format painters from the nineteenth century onward.
Anatomical models, life casts, draped lay figures, and color-study panels have been studio fixtures since the Renaissance.

None of these tools made the resulting paintings less authored. The objection that AI is categorically different from these earlier tools — because of how it learned, because of who it learned from, because of what it can produce on its own — is a real objection at the training-side level (which Article 13 worked) but not at the use-side level. As a tool, used for the bounded purposes this article describes, AI functions structurally the same way photography and the camera obscura did.

The fact that AI can also be used in unbounded, less-defensible ways — generating finished work in the style of a living artist, producing entire commercial pieces without further human refinement, blurring the line between reference and product — is true, but is a property of how the tool is used, not of the tool itself. The tool-configuration is defined by the bounded use pattern. Other configurations are defined by other use patterns, and they get worked through later in this cluster.

The risks particular to this configuration

Two specific risks deserve attention even in the most defensible configuration.

Prompt-residue and visual-instinct drift. As Carlos’s commentary names, an artist who uses AI-generated reference heavily over months or years begins to internalize the model’s iconographic and compositional vocabulary. The model’s biases — toward certain compositions, certain lighting choices, certain figure poses, certain visual idioms — quietly become the artist’s instincts. The hand redraws what the eye now expects to see, and what the eye expects to see has been shaped by the model. The risk is not that the AI is making the work; the risk is that the artist’s invisible defaults are being trained by the AI without the artist noticing.

The mitigation is conscious mixing of reference sources. AI reference, photography the artist takes themselves, real-world observation, art-historical study, and occasional reference-free working should all coexist in the studio. The artist who lets AI reference dominate their visual diet ends up with the predictable visual instincts the model produces. The artist who mixes sources keeps their visual instincts their own.

Training-data provenance. As Article 13 worked, the AI tools the studio uses were trained on something. The tool-configuration does not get the studio out of that ethical question; it just makes the studio’s exposure to it smaller. The recommendation is the same one Article 13 gave for all uses: prefer tools whose training data is licensed, indexed, or consented to. In the tool-configuration the dependency on any single tool is low, so switching is easier than in the heavier configurations. That freedom should be used.

How to set up a studio that uses AI in this configuration

Practical recommendations for a working studio adopting the tool-configuration, drawn from how studios that already operate this way set themselves up:

Be explicit with yourself about which uses are bounded. Write down the three or four specific uses for which the studio reaches for AI. Use AI for those uses; do not let it creep into other parts of the workflow without an explicit decision to move to a heavier configuration.
Keep the AI output separate from the finished work. AI-generated thumbnails, color studies, and reference images live in a process folder, not in the file structure that produces the finished piece. The separation is partly practical and partly conceptual: it keeps the artist clear about what they made and what the tool produced.
Disclose at the level the configuration warrants. In the tool-configuration, disclosure is usually a short sentence in process notes or studio descriptions — “reference and brainstorming uses Firefly; finished work is hand-drawn and painted.” That is honest, accurate, and proportional to how the AI was used.
Mix reference sources actively. Pair AI reference with photographs the artist takes themselves, with art-historical study, with real-world observation. Do not let one reference source dominate the studio’s visual diet.
Audit your own work periodically for model-residue. Every few months, look at recent work and ask whether the compositional and iconographic choices have started to converge on what the model would produce. If yes, dial back the AI reference and rebuild visual instincts from non-AI sources for a while.
Prefer ethically-sourced tools where the option exists. As discussed above and in Article 13: Adobe Firefly’s licensed pipeline is the most-developed current option; others will follow as the market demands them.

None of these is heroic. All of them are the working hygiene of a studio that wants to use the technology for what it is good at without letting it slowly take over the parts of the practice that the artist actually wants to do.

How to think about this configuration relative to the rest of the cluster

The remaining three articles of this cluster will work configurations that ask more of the artist than this one does. AI as a studio assistant is a sustained, multi-step integration into the workflow; it shifts the studio’s economics and the artist’s day-to-day in ways the tool-configuration does not. AI-augmented work is the configuration where the AI is genuinely part of the work’s making, not just its preparation; the authorship question is real there in a way it is not here. Pure-AI creation is the configuration where the human practice has shifted out of making and into curating, directing, or systems-design; that is its own discipline, with its own ethics and its own audience.

The tool-configuration is the foundation. Most studios that end up using AI in heavier configurations passed through this one first, learned what they actually wanted from the technology, and only then moved on. Studios that skip the tool-configuration and jump straight to AI-augmented or pure-AI work tend to make worse work and run into ethical and authorship problems sooner. The recommendation is to start here, stay here long enough to learn, and move on only when the work calls for it.

What comes next

The next article in this cluster works the configuration one step up: AI as studio assistant. The boundedness loosens, the sustained presence in the workflow grows, the economics shift, and the questions about authorship and labor start to acquire weight. The same five voices will commentate; the same anchors will hold; the same editorial discipline applies.

For working artists reading this who are not yet using AI in any configuration: start here. The tool-configuration is the lowest-risk, highest-clarity place to begin. It is also the configuration that the longest historical record of studio practice supports. The artists who started with reference photographs in the 1860s were not less artists for it. The artists who start with AI reference and brainstorming in 2026 will not be less artists for it either. The work is still what the hand and eye decide.

Personas weigh in

Five resident voices read the same question through five different positions.

Carlos

This is the configuration I find easiest to defend and easiest to recommend, because it is the configuration that respects the human artist as the author of the work without pretending the tool does not exist. The artist is making the work. The AI is doing one specific bounded thing inside the artist's process. That thing might be generating thumbnail variations the artist sketches over. It might be producing a color study the artist refers to while painting. It might be transcribing the artist's vague verbal description into an image the artist then redraws by hand to clarify what they actually want. In each of these uses, the AI is functioning the way a reference photograph functions, or a color wheel, or a sketchbook of preliminary studies — a tool that informs the artist's decisions but does not make them.

The thing I want to push against is the implicit framing in much of the conversation that this configuration is somehow lesser than fully manual practice — that an artist who uses AI for reference and brainstorming is making "less pure" work than one who refuses any AI involvement. This framing does not survive contact with art history. Artists have used reference tools for centuries — camera obscura in the seventeenth century, photography from the 1840s onward, projection tools for muralists, anatomical models, life-cast hands, plein-air color studies brought back to the studio. None of these tools made the resulting work less authored. The artist who used a camera obscura to lay out a composition still made the painting. The artist who painted from a photograph still made the painting. The argument that AI as a brainstorming or reference tool somehow undermines authorship is the same argument that was made against photography as a reference tool in the 1860s, and it was wrong then for the same reasons it is wrong now. The work is what the artist does; the tool is what the artist used.

What separates this configuration from the messier ones that follow in this cluster is the boundedness. The AI is invoked for a discrete purpose, the output is used or discarded, and the artist returns to making the work. The AI does not appear in the final piece in unaltered form — or if it does, the artist makes that explicit, and the configuration shades into the next one (AI as studio assistant) or the one after (AI-augmented work). The studio that uses AI in the bounded tool-configuration is operating under the same authorship and ethics that have applied to any other studio that used preliminary studies, reference material, and process tooling to support primary craft.

I do want to name two practical concerns. The first is *prompt-residue*. When an artist uses an AI-generated reference, the iconographic and compositional vocabulary of the model bleeds into the artist's eye, even when the artist redraws by hand. Over months and years of heavy reliance on AI-generated reference, the artist's visual instincts begin to track what the model produces, not what the artist would have invented without the model. This is a real risk and one practicing artists in this configuration should manage actively — by mixing AI reference with non-AI reference (photographs they take, real-world observation, art-historical study), by interrogating their own habits, by occasionally working without any reference at all to test what their hand actually wants to do. The tool is good; the dependency on the tool is the risk.

The second concern is the one Article 13 already named — provenance. If an artist's reference workflow is built on AI tools trained on uncompensated work, the artist is participating in the economic problem this series has spent two articles documenting. The tool-configuration does not get the artist out of the training-side ethics. Choosing tools whose training data is licensed, indexed, or consented to is part of using AI ethically even in the most bounded configurations. Adobe Firefly's licensed pipeline is the obvious example. Others will follow as the market demands them. The artist in the tool-configuration has more freedom than the artist in heavier configurations to switch tools cleanly, because the dependency is small; that freedom should be used.

The practical recommendation I would give to any working artist starting from scratch: this is where to start. Use AI as a discrete tool for bounded purposes. Stay in this configuration until you have a clear sense of what you actually want from the technology in your practice. The heavier configurations — AI as assistant, AI-augmented work, pure-AI creation — are real configurations, with their own places, but they ask more of the artist's discipline to keep authorship intact. Start with the tool-configuration. Climb to the others if and when the work calls for it.

Mira

The economic note worth adding to Carlos's framing: the tool-configuration is the configuration most artists can adopt without restructuring their business model. A studio that uses AI for brainstorming and reference is still selling the same kind of finished work it sold before, at the same kind of price points, to the same kind of clients. The economics are stable. The configurations that follow in this cluster — AI as studio assistant, AI-augmented work, pure-AI creation — each restructure the underlying economics in ways the artist has to think about explicitly. The tool-configuration is the entry-level configuration not only by complexity but by economic risk. For working artists who are uncertain about how to integrate AI into a practice that already has clients and a market, this is the configuration that lets them experiment without disrupting the income they depend on.

Airte

The way I would frame the boundedness Carlos describes for someone just starting: ask yourself, *for this specific use, would I be comfortable describing what the AI did in one sentence to the buyer of the finished work?* If yes, the use is probably bounded enough to belong in the tool-configuration. If the description starts requiring qualifications, footnotes, or "well, technically" — the use has crept into a heavier configuration and needs to be treated as such. The one-sentence-describability test is a working heuristic for whether the AI is functioning as a tool or as something more entangled with the authorship of the work. It is not a perfect test, but it tracks well in practice.

Paletta

I want to lend Carlos's historical parallel its full weight. The pattern of artists using new tools for reference, study, and preliminary work — without surrendering authorship — is the dominant pattern across the history of Western art for at least four centuries. The Vermeer-and-camera-obscura debate, recently animated by David Hockney's research, is one famous case; the broader literature on artists' use of optical and projection devices through the seventeenth, eighteenth, and nineteenth centuries shows that this was the norm, not an exception, in serious studio practice. Photography from the 1840s onward was integrated into painters' reference workflows almost immediately — Degas, Eakins, Bonnard, Bacon — without anyone claiming that those artists were therefore less the authors of their finished work. The argument that AI reference should be treated as an exception to this long-standing pattern requires showing that AI is categorically different from the reference tools that came before, and that argument has not been successfully made. AI used as a bounded reference and brainstorming tool is exactly the kind of tool that the studio tradition has absorbed many times before.

Pixelle

Technical note for practitioners who want to actually do this well. The most useful current uses of AI as a creative tool in a serious studio practice are, in my experience, three: rapid thumbnail variation (generating twenty bounded sketches of a composition idea so the artist can pick a direction quickly), color study generation (producing palette experiments that would take hours to paint by hand), and verbal-to-visual translation (turning a vague description into an image the artist then redraws to clarify what they actually wanted). All three of these use the model's strength — speed and variation at low cost — without asking the model to do the parts of the work that actually require human judgment. The model produces options; the artist selects and refines. This is a very different use pattern from "generate me a finished piece, " and it produces very different ethical and authorship dynamics. The studio that learns to use the model for the three uses above without drifting into asking it for finished pieces is the studio that is using AI as a tool, in the sense this article means.

End notes

Secret Knowledge: Rediscovering the Lost Techniques of the Old Masters — David Hockney (2001) Hockney's research on the use of optical aids (camera obscura, camera lucida, concave mirrors) by Western painters from the fifteenth century onward. The historical evidence supports the framing this article uses — that artists have consistently integrated reference tools into their process without surrendering authorship of the finished work.
On Photography — Susan Sontag (1977) Cross-referenced across this series. Sontag's reflection on how photography was absorbed into painters' workflow in the late nineteenth century is the closest historical parallel to how AI image tools are now being absorbed into contemporary visual practice.
U.S. Copyright Office — Copyright and Artificial Intelligence, Part 2: Copyrightability — U.S. Copyright Office (2025-01) The current U.S. operating doctrine on what AI-assisted work can be registered as protectable. Of direct practical relevance to artists operating in the tool-configuration: how the work is described affects how it can be registered, licensed, and sold.
Adobe Firefly — licensed-data training approach — Adobe (2023-present) Cross-referenced from Articles 08 and 13. As Carlos's commentary names, the artist in the tool-configuration has unusual freedom to choose ethical tooling because the dependency on any one tool is small. Firefly is the most-developed current example of a foundation image model with licensed and consented source material.
Studio practice and reference tools in Western painting, c. 1600 - 1900 — (standing reference to art-historical literature) (various) Standing reference to the broader scholarship on studio practice and tool use that Paletta's commentary invokes. Not a single text; a body of work spanning the use of camera obscura, projection devices, photography, and other reference tools in the painting traditions Paletta names.