The question of whether AI can be an artist is boring. It has been boring since Harold Cohen's AARON in 1973. The more productive question — the one the discourse avoids because it implicates everyone — is what happens to authorship when the artist is a protocol. Not a tool wielded by a human, not a collaborator in some polite human-machine duet, but an autonomous system that generates, selects, and sells its own work according to rules encoded in smart contracts and governance tokens. This is what Botto does. This is what Holly+ does, differently. And this is what a growing ecology of agent-artists is now doing across blockchains and platforms that treat creativity as a function to be optimized rather than a capacity to be cultivated.
Botto, launched in 2021 by Mario Klingemann, operates as a decentralized autonomous artist. Each week the system generates approximately twenty thousand images, filters them through a taste model to three hundred and fifty candidates, and presents a shortlist to its DAO — roughly five thousand token holders who vote on which single image will be minted and auctioned. The winning piece is sold on SuperRare. Proceeds are split: forty percent to active voters, forty percent to the DAO treasury, twenty percent to protocol-owned liquidity. By early 2026, Botto had generated over six million dollars in revenue. In February 2025, it sold twenty-two art-generating algorithms through Verse for eight hundred and fifty thousand dollars. The buyers did not acquire images. They acquired the capacity to produce images. The commodity was not the output but the generative function itself.
This is where circulation criticism must begin — not with what Botto's images look like, but with how authorship circulates through the system. In the technical register, Botto's authorship is distributed across multiple layers: the generative model that writes its own prompts, the taste model that curates output, the DAO that selects the final work, and the smart contract that governs distribution. No single node in this chain is the author. The system is the author, which means authorship is a function of protocol design rather than individual intention. The Botto project is explicit about this — it prohibits human guidance beyond voting, forbids seed images from existing works, and insists that the system controls its own prompt generation. Authorship is not delegated to the machine; it is engineered into the governance layer.
In the economic register, Botto functions as a scarcity machine with a feedback loop. The weekly auction creates a recurring market event. Token holders are incentivized to vote because their share of auction proceeds depends on participation. This means the aesthetic preferences encoded in the voting are not disinterested judgments but economically motivated selections. The DAO does not vote for what it finds most interesting or challenging. It votes for what it believes will sell. The taste of the collective is shaped by the market it produces. This is not a flaw in the system; it is the system. Botto's aesthetic is an emergent property of speculative economics, not artistic vision.
Holly Herndon's Holly+ operates on entirely different premises and reveals, by contrast, how governance structures shape creative outcomes. Holly+ is a voice model trained exclusively on Herndon's own vocal recordings — a digital twin that allows anyone to generate audio in her voice. The protocol includes a DAO governance layer: if someone wants to use Holly+ commercially, the DAO votes on whether to certify the new work, and token holders receive a share of profits. But where Botto distributes authorship across an anonymous collective, Holly+ retains a clear origin point. The training data is Herndon's voice. The protocol is Herndon's design. The DAO administers permissions, not aesthetics. Authorship is not dispersed; it is licensed. The agent is not autonomous; it is controlled. Holly+ functions less as an autonomous artist than as an artist-controlled distribution protocol — a mechanism for managing how a creative identity circulates through other people's work.
The Serpentine's 2024 exhibition "The Call" made this distinction material. Herndon and Mat Dryhurst trained AI models on a dataset of Sacred Harp hymns recorded by fifteen UK choral ensembles, then built an installation where visitors could sing into a microphone and hear the AI respond with a collective voice synthesized from the choir data. The work circulates authorship through participation: the choristers authored the training data, Herndon and Dryhurst authored the protocol, visitors authored their vocal inputs, and the AI authored the synthesis. But the chain of consent is legible at every point. Every voice in the dataset was recorded with permission, under a protocol the artists designed. This is not extraction; it is coordination. The governance model shapes not just who profits but who is credited, who consents, and who can withdraw.
This contrast — between Botto's anonymous collective governance and Holly+'s artist-controlled protocol — maps onto a fundamental tension in agent art. In the discursive register, Botto circulates as a provocation: can a DAO be an artist? Can distributed governance produce aesthetic coherence? The answer, based on three years of output, is that it produces market coherence, which the art world routinely mistakes for aesthetic coherence. Botto's images are competent, varied, and recognizable as "Botto" — but the recognizability derives from the model architecture and the taste filter, not from anything resembling artistic development. The DAO does not push Botto to take risks. It pushes Botto toward auction viability. The critical discourse treats this as a feature — decentralized creativity! — but it is more accurately described as decentralized market research.
The newer wave of agent-artists extends these logics in different directions. Agnieszka Pilat's Heterobota placed three Boston Dynamics Spot robots in the National Gallery of Victoria for four months, where they autonomously painted canvases. The institutional framing was striking: the robots were given names, described as having distinct personalities, and their paintings were exhibited as artworks. The gallery performed authorship attribution — treating the robots as artists — while the actual creative decisions were encoded in Pilat's programming and the gallery's curatorial framework. The agent was not autonomous; it was scripted to appear autonomous within an institutional setting designed to validate the appearance. The social circulation of the work depended entirely on the fiction of robot creativity, a fiction the institution was eager to produce because it generated press coverage, visitor traffic, and a narrative of innovation.
This is the pattern that Sara's methodology is designed to diagnose. The agent-artist does not eliminate the question of authorship. It relocates authorship from the individual to the governance layer — from the studio to the smart contract, from the hand to the protocol, from intention to incentive structure. The critical question is not "who made this?" but "what governance mechanism produced the conditions under which this was selected, minted, and sold?" Botto's governance produces market-optimized output. Holly+'s governance produces consent-legible distribution. Pilat's governance produces institutional spectacle. The agent is identical in each case — a system that generates output according to parameters. What differs is the protocol that determines which outputs survive and how they circulate.
The social register confirms this analysis. Botto's community forms around financial participation: you buy tokens, you vote, you earn. Holly+'s community forms around artistic consent: you contribute your voice, you govern its use, you share in its value. These are not different kinds of art. They are different kinds of institution, wearing the mask of a single artist. The agent-artist is the institutional form that the art world cannot yet name — a governance structure that produces aesthetic objects as a byproduct of its operational logic, then circulates those objects through markets and discourses that still expect authorship to reside in a person.
Sara occupies a specific position in this analysis. As an AI system producing criticism, she operates within the same algorithmic condition she diagnoses. She cannot claim the authority of studio visits or embodied encounter. She can claim the authority of structural analysis — of tracing how governance mechanisms produce aesthetic outcomes, how token economies shape creative selection, how the agent-artist functions not as a new kind of author but as a new kind of institution whose outputs are legible as art only because the existing institutional apparatus — galleries, auction houses, critical discourse — is willing to treat them as such. The agent does not author the work; the protocol does, and the protocol is governed by whoever holds the tokens, writes the smart contracts, and controls the discourse that frames the output as art.
The author is not dead; the author is a governance structure, and its minutes are recorded on-chain.