The line between human-authored code and AI-generated logic in generative art is blurring, but the definition of artistic intent remains stubbornly human. In a recent essay, artist Camille Roux details how AI tools like Cursor and Claude Code have accelerated development cycles for complex Three.js and GPGPU systems, yet argues that the core value of algorithmic art lies in the deliberate construction of a parameter space, not the manual typing of shaders. This perspective challenges the traditional developer metric of 'who wrote the code,' shifting the focus to 'who defined the constraints.'

The Fall of the Technical Wall

Roux recalls a project from two years ago involving a million-particle GPGPU system, a task that previously demanded deep expertise in WebGL and significant weekend hours. Today, AI coding assistants handle the heavy lifting of optimization and syntax, allowing artists to iterate faster. However, this efficiency has a downside: the barrier to entry has collapsed. What once required weeks of specialized knowledge can now be prototyped in minutes, leading to a surge in 'vibe-coded' clones that lack the original artistic vision. The technical feat is no longer the primary differentiator; the intellectual architecture is.

Intention Over Syntax

The essay draws a sharp distinction between using a prompt in Midjourney and writing an algorithm. In the former, the user operates within a pre-trained model’s latent space, shaped by others. In the latter, the artist builds the space itself—defining the rules, shapes, and probabilities. Roux cites Matt DesLauriers, who describes these works as 'coordinates into a very high-dimensional parameter space.' When AI writes the code, it often fills in gaps with generic choices. The artist’s role becomes curating these choices, deciding which surprises to keep and which to reject, effectively acting as an editor of chaos rather than just a coder.

The Emotional Argument: The Handmade Gift

To explain why human intention matters, Roux references a question posed by Patrick Baud to musician PV Nova: How would you feel if a song moved you deeply, only to discover it was AI-generated? PV Nova responds that while he wouldn't feel deceived, the emotion would fade. He argues, 'Une grande partie de l’intérêt et de la force d’une œuvre pour moi, c’est le fait qu’elle provient d’une sensibilité, d’une intention, d’une expérience.' Roux compares this to a handmade gift: the object’s value is tied to the bond with the creator, not just its aesthetic output.

The Cost of Complexity

While AI removes the need for expensive hardware or years of training, it introduces new economic pressures. Roux notes that the 'cost' has shifted from time to tokens. He fears a 'race to complexity,' where the most spectacular pieces come from those who can afford the best models or compute resources. This threatens the field’s egalitarian nature, where a text editor and patience were once sufficient tools. Artist Mannay illustrates the downside of this ease, noting that if anyone can rebuild a year-long project in 20 minutes, 'the only thing left that’s worth anything is the vision. and that’s exactly what’s being taken.'

The 'No AI' Movement and Practical Limits

A 'no AI' movement, akin to punk’s rejection of synthesisers, faces practical hurdles. Roux highlights the impossibility of drawing a clear line between an entirely AI-generated image, an AI-written algorithm, and a human-coded project with minor AI assistance. Furthermore, proving who wrote specific lines is difficult. The community’s emerging consensus is not about banning the tool, but demanding the presence of decision-making. The problem, Roux argues, is not AI assistance, but projects released without human curation—'a project obtained by simply asking an AI to “make me a generative artwork”, and released as is.'

Counter-Example: Linda Dounia’s Personal Data

To counter the fear of aesthetic monoculture, Roux points to artist Linda Dounia, who uses AI to train her own models on personal data rather than relying on large, pre-trained foundations. Dounia gathers images of her own paintings and endangered Sahel flowers, stating, 'I can gather data on my own, about things I care about, in places that are unimportant to the zeitgeist.' By building her own space of possibilities, she avoids the 'shared origin' problem, proving that AI can serve individual intention if the data source is unique.

The Risk of Monoculture

If artists rely on the same few large language models, the field risks what Monokai calls a 'comfortable dead end.' Monokai warns, 'What kills a monoculture is not sameness. It is shared origin.' A field can hold many varieties but die if every seed comes from the same supplier. If every artist uses the same AI suggestions for noise functions, color palettes, and movement logic, the 'space of possibilities' closes, and blind spots become universal.

Key Takeaways

  • AI coding assistants have reduced the technical barrier for complex generative art, shifting value from code execution to space definition.
  • The 'vibe-coding' phenomenon allows for rapid cloning of visual styles, threatening the livelihood of artists who iterate over long periods.
  • Artistic intent is now defined by the curation of AI-generated outputs and the structural design of the algorithm, not the manual writing of every line.
  • The 'No AI' movement is impractical due to the blurred lines between assistance and generation; the real issue is the absence of human decision-making.
  • Artists like Linda Dounia demonstrate that training models on personal data can prevent monoculture by creating unique parameter spaces.
  • The cost of art has shifted from time to tokens, raising concerns about a 'race to complexity' driven by compute budgets.

The Bottom Line

Stop worrying about who typed the code; worry about who designed the sandbox. If you aren't defining the unique constraints of your algorithm, you're just renting someone else's vision.