At Tesla’s October 2024 unveiling of its vision for an autonomous future, called “We, Robot,” several of the company’s sleek humanoid “Optimus” robots meandered through a crowd, served cocktails and hors d’oeuvres, and mingled with guests. This enactment of an industrial-era dream of automated labor celebrated embodied artificial intelligence. Within days, as is often the case with supposedly autonomous technologies, news broke that the robots’ movements and quick, contextually apt verbal replies were the work of human operators.1 The merging of a technical demo with cyber-utopian spectacle highlights how AI autonomy is as much about appearance, form, and style as it is about so-called autonomous capacity. A few months later, a smaller, more intimate theater of the same promise of automated intelligence was offered by the “Friend” pendant, an always-listening AI companion worn as a necklace. This time it was framed as a specific type of work: companionship and presence made possible through a low-friction intimacy with computation.2 “In 2023, after dropping out of Harvard, [Friend’s founder Avi] Schiffmann started raising money to build an artificially intelligent ‘wearable mom.’”3

What distinguishes these technologies from earlier forms of automation? What is behind their potency in capturing the desirous investment of users? How can we understand what the aesthetics of AI—which we might call the autonomous sublime—conjure, even as we witness the future performed by tech billionaires fail to arrive? And most importantly for a feminist analysis, how are users’ embodiment and affect captured by these technologies?

As Walter Benjamin noted, in the age of mechanical reproduction, the original work of art loses its aura.4 We wish to update and tweak Benjamin’s argument for the age of AI aesthetics. Despite the capacity of generative models to infinitely reproduce language and images, we argue that AI aesthetics can actually be best understood as the aesthetics of the autonomous sublime combined with ubiquity, regimentation, and automation. We call this AURA: Aesthetic, Ubiquitous, Regimented Automation. Building on Benjamin’s observation that the political project of fascist aesthetics is to organize subjects into a mass audience of users, we assert that AI AURA is designed to capture human intimacy through the affective and emotional engagement of embodied response. Thus, AURA serves to name the production of AI output in the complete absence of an original, even as this production is dependent on an extraction of style.

AI aesthetics is the conversion of culture into an operational resource. Here, the aesthetic capture of affect as tone, intimacy, vibe, and the engineered feeling of connection makes delegation and surveillance livable. This capture is ubiquitous insofar as AI is an ambient infrastructure of everyday life, dispersed across feeds, prompts, cameras, microphones, recommendation stacks, workplace dashboards, and automated “assistants.” It is regimented in the sense that AI’s ambient infrastructures are governed by platform monopolies, intellectual property enclosures, terms-of-service politics, and the social sorting that turns legibility into control. Moreover, the automation undertaken by AI is not mobilized to reorganize production; rather, production reorganizes aesthetics under monopoly capitalism. This is style extracted and turned into computable features. Taken together, AI’s AURA turns the social texture of style, speech, and intimacy into a scalable apparatus of capture and recombination, thus making these elements alienable as both labor and property.

As we argued in our 2019 book, Surrogate Humanity: Race, Robots, and the Politics of Technological Futures, fantasies of robotic and AI futures, whether framed as liberation from labor or as neutral progress, are bound to political projects that reproduce the dispossessions of racialized capitalism and empire, even as some of them claim to transcend them. This matters for grappling with generative AI, because its AURAtic qualities can be read as forms of infrastructure on the one hand, and fascist techniques of control on the other.

If Marx’s theory of alienation under industrial capitalism can be glossed as the entire human sensorium being replaced by the sense of “to have,” and as all of our social relations being performed by commodities as our surrogates, then what is the new sensory apparatus that accompanies acceleration in the age of AI?5 Alienation extends from the realm of labor into all of life, while capitalism sells ways of being, separating us even from our own desires. Thus, political life becomes staged and aestheticized. In The Dialectic of Enlightenment, Horkheimer and Adorno underline the way of fascist propaganda bypasses persuasion and goes directly for libidinal attachment, harnessing affective sensibilities like fear and identification. The ultimate goal of this libidinal attachment is for people to obey. Aesthetics is used to bypass rational thought and provoke a directly physical and emotional response. Fascist aesthetics produce obedience through form, promising authenticity while enforcing sameness.

In this essay we move across a variety of AI agents (that we think of as artefacts) to consider how these highly stylized enactments of AI AURAs are used by tech corporations to iterate what AI is, how it should be felt, and why it should be allowed to reorganize life and human desires. Thus, the question is what happens when we are forced to encounter those enactments, not as just consumer products or accessories but as aestheticized, operational, affective, and stylized ubiquity. In the 1920s, Soviet avant-garde tactics of operativism named the modality through which images and texts become a mode of action. Today, AI does not just perform functions for the sake of efficiency; AI systems are designed to persuade, organize, dazzle, and enchant, and thus to capture users’ affective investment.6

Benjamin’s diagnosis of fascist aesthetics helps us to grapple with why this conversion entails both pleasure and participation. Writing about the reproducibility of art in a different moment of technological transformation, he stated:

That which withers in the age of mechanical reproduction is the aura of the work of art. This is a symptomatic process whose significance points beyond the realm of art … detach[ing] the reproduced object from the domain of tradition. By making many reproductions it substitutes a plurality of copies for a unique existence. And in permitting the reproduction to meet the beholder or listener in his own particular situation, it reactivates the object reproduced. Both processes are intimately connected with the contemporary mass movements.7

In identifying the AURA of AI, we aim to describe a power grab by a small group of elite actors who bypass existing forms of governance, and even the social sphere, to organize users through the direct capture of affective investment. As Matteo Pasquineli has written, “Technical artifacts reveal the form of the society that surrounds and runs them.”8 Today, such artifacts illustrate AI’s AURA potential for the consolidation of fascism, as they conceal modes of reproductive labor and the reproduction of style and substance in generative AI, whether we read this reproduction as theft or extraction. The “aesthetics of AI” is not reducible to the familiar debates about whether generated images can fool people into thinking they are real or fake, whether Studio Ghibli’s style has been accurately reproduced by ChatGPT, or whether chatbots “sound” or “feel” human. It is rather the ensemble of forms (sleek or cute robots, intimate wearables, servile or seductive voices, and pleasurable interfaces) through which monopoly platforms secure the right to capture, recombine, and operationalize life. As the artefacts and aesthetics of AI become an instrument of extraction, the myth of automation is embraced and accepted because it looks, sounds, and sits on and in the body like the only possible future.

1. Optimus: Captivating Automation

Niamh Ancell, “Tesla Optimus bots actually controlled by humans during ‘We, Robot’ event,” Cybernews, October 15, 2024,

Tesla’s Optimus robot is slated to be available for purchase for $30,000 in 2027. It is designed to perform unsafe, repetitive, and boring tasks. In the image above, we see Optimus washing dishes. This still encapsulates the fantasy that historically unpaid housework, the work of reproductive labor, will be taken over by robots in order to free humans for more valued tasks that supposedly require more “intelligence.”9 Techniques of computational control secure consent for automation by displacing attention away from the surrogate worker. As a result we are largely blind to things like human labor in data labeling and teleoperation, resource extraction and logistics, legal enclosures around models and platforms, and the institutional asymmetries that decide which lives are rendered as training material and which lives are protected as property. This is not just about labor, but also about the aesthetics of AI. Even if a system is partially controlled by a puppeteer, the interface can still present the interaction between user and tech as seamless, immediate, and machine authored. The user experiences a unified persona (“the robot,” “the assistant”), even when the performance is stitched together by remote operators and tech boosters. It is a production technique for the appearance of agency. Tesla has been openly hiring “Data Collection Operators” to wear motion-capture suits and VR equipment while performing the gestures the robot is supposed to learn. They walk for hours, repeating movements, becoming, in effect, a human prosthesis for the machine’s future.10 Teleoperation is not an embarrassing revelation that unmasks the AI theater; it is rather part and parcel of the mechanism by which the “automated future” is made inevitable in the present.

The We Robot event showcased the bipedal robots dancing in glass cages. The allusion to the gendered and racialized spectacle of caged performers can be juxtaposed with promotional images of Optimus in the domestic space, washing dishes and unloading groceries. But just like Optimus can’t dance without human intervention, it also can’t currently wash dishes. This fantasy, and the juxtaposition of dishwashing and caged dancing, is important for illustrating one of the core arguments about the racial and gendered figuration and fantasy of freedom. In her book Scenes of Subjection, Saidiya Hartman discusses the concept of the surrogate in the context of US chattel slavery. Hartman brings up novelist Toni Morrison’s formulation of slaves as “surrogate selves for the meditation on the problems of human freedom.”11 Today, even as more valued tasks within capitalist regimes of production and accumulation, such as knowledge work, become automatable, the stated goal of technological innovation is still to liberate human creative potential. Yet when it comes to what activities require human intelligence, the goal posts keep shifting.

Feminist materialist accounts of social reproduction argue that the production of surplus value doesn’t depend on waged production alone. It also relies on vast amounts of reproductive labor, including care, cleaning, cooking, child-rearing, emotional maintenance, and the day-to-day work of sustaining workers and social life. As scholars like Silvia Federici, Angela Davis, and others have argued, this work has historically been gendered, racialized, and often unwaged or underpaid.12 If we understand “the household,” “care,” and “affect” as not external to the economy but as its enabling infrastructure, then we can analyze Optimus’s AURA as it straddles the cage and the dishwasher. Its humanoid form bridges automation and the intimate space of the home, concealing the property relations through which labor is rerouted on AI’s monopolistic platforms. As an artefact, Optimus could be intended for grander things, but this history of devalued tasks is embedded in its infrastructure and its affective engagement. The humanoid form and the service-performance aesthetic import the cultural grammar of care and servitude, making exploitation feel like convenience, without troubling the underlying gendered and racialized division of labor. Optimus does not emancipate reproductive labor but rather extracts its forms (gestures, tones, routines) into a computable “style,” redeploying them as automated capacity.13

2. Friend: Monetizing Mom Vibes

Photo illustration by Derek Brahney.

The Friend pendant compresses the same logic into a tiny accessory. Friend is a necklace that routes ambient life into a companion app and offers a running commentary. It is meant to be a supportive presence but is often experienced as socially abrasive, intrusive, or simply alienating.14 Described as “a chipper white pendant that, for $129, hangs around your neck, ‘always listening,’ ready to quip and comment on the IRL goings-on of your life,” the aesthetics of Friend include not only the artefact’s minimalist charm, but its promise of closeness without reciprocity, and companionship without social obligation.15 The device’s core promise—continuous attention to scenes of everyday life, from the office to happy hour—doesn’t just commodify care but automates it. The pendant is an interface that normalizes capture so that interaction itself can be formatted for computational processing. In other words, through the intimacy of 24/7 presence, life becomes data and data becomes the substrate for further automating relations—restaurant recommendations, persuasion around purchasing, scheduling, policing the self, and even affect modulation. The aesthetics of AI’s autonomous sublime operates here through sensing: sensing the wearer’s emotional state of mind, their context, and their need for advice. AI artefacts are the soft power that makes automation livable and desirable. Whereas Optimus is designed to have the function and presence of servitude, Friend captures affect through trust and proximity.

The AI-engineered experience of style centers ownership, which is integral to the AI revolution. Consumer AI’s accessible, on-demand creativity, friendly agents, and mom proxies index the aestheticization of present-day monopoly capitalism. In the epilogue to “The Work of Art in the Age of Mechanical Reproduction,” Walter Benjamin argues that fascism “attempts to organize the newly proletarianized masses without affecting the property structure.” It does this by giving them “not their right, but instead a chance to express themselves … The logical result … is the introduction of aesthetics into political life.”16 Nearly a century later, Friend’s sensory style and approximation of intimacy still prioritize ownership, but with much greater degree of consolidation (model ownership, platform enclosure, data extraction, and compute concentration). In Benjamin’s terms, the aesthetic does political work: it organizes subjects into a mass audience of users, turns technical power into an affective spectacle, and rebrands dispossession as empowerment.

At the end of summer 2025, Friend ran a large New York City subway campaign, plastering ten thousand posters across trains and stations. Reported to have cost roughly $1 million, it caused public backlash.17 Within days, many of the ads were torn down or defaced. Commuters and activists scrawled messages over the ads about the dangers of surveillance and the importance of real friendship. “AI is not your friend,” read one message. Images of the altered ads circulated widely online. The main objections to Friend thus converged around opposition to the packaging of surveillance as companionship; in effect, they regarded the campaign as an attempt to enclose the affective commons.18 Friend’s always-on microphone, recording everything uttered in both private and public spaces, was particularly reviled for its affective manipulation and its promotion of dependency. On top of all this, Friend was also a faulty product that didn’t work as advertised.19 We can read this as an interruption of the sleek, fascist AI aesthetics of inevitability. Read alongside Benjamin, the defacing of the Friend ads contests the aestheticization of a politics that not only leaves property relations unchanged, but extends them so that the objects we own now own us—or more precisely, our data.

3. Scarlett Johansson’s Voice: Alluring LLMs

In 2024, OpenAI asked the actor Scarlett Johansson if the company could use her voice for their soon-to-be released voice assistant. She turned them down. But when the assistant, called “Sky,” was released a few months later, the actor said that its voice was “eerily similar” to her own.20 Though representatives of OpenAI claimed that another voice actor was used for the data informing Sky’s “voice,” they soon pulled the assistant after increased scrutiny. At issue was how to define “voice” in terms of defensible intellectual property, as well as more broadly, how AI should be regulated. Is a voice that “sounds like” that of an identifiable individual connected to their legal rights of privacy and property? In the case of a celebrity, the relevant right is the “right to publicity,” which affects people whose identity is their livelihood.21

While the “theft” of voice (or what we have been calling the extraction of “style”), as well as the AI use of other intellectual and creative content without authorial consent, has emerged as a new and uncharted legal issue, the juridical solution hinges on a reassertion of the right to privacy and intellectual property. Rather than pit one property right against another (which aspect of capitalist ownership will win the day?), it is critical to consider how and why the voices of AI assistants deploy certain aesthetics. Johansson’s objection, framed through the logic of the right of publicity and the voice as property, is therefore not only about one celebrity’s voice; it also marks the frontier where the aesthetics that smooth AI’s entry into everyday life collide with the property claims and monopoly incentives that drive scaling in the first place.

AI interfaces, especially those incorporating voices, trade on an aesthetic of intimacy (comforting, servile, seductive, companionable) to make a probabilistic system feel like a person, a helper, a presence, or an object of desire. The film Her features Scarlett Johansson herself as the voice of an AI companion that the male protagonist falls in love with. The use of celebrity-adjacent voices for AI assistants is the most concentrated form of this aesthetic strategy, offering a shortcut to trust through recognizability and cultural memory. But this familiarity is also a vector for enclosure, since if a platform can appropriate (or closely approximate) the timbre, cadence, and affective signature of a voice, it is effectively converting aesthetic labor into a reusable, automatable interface. Style becomes operational capacity.

The way that technologies make us “feel human” depends on AURA. Part of their seduction is the experience of being attended to by a particular kind of body that we imagine behind the voice. However, the aesthetics of automation at work in these artefacts are not about authenticity. They are about the emotional or pre-emotional affective responses provoked in the body of the user. In his essay “The Grain of the Voice,” Roland Barthes theorizes the semiotic and political potential of the embodied response of one human to the voice of another. Barthes is trying to parse the relationship between language and semiotic systems. The grain of the voice does something to language that can be referenced by adjectives, but the grain also causes our bodies to react in a way that is “erotic,” sensual.22 This causes us to lose the subject. It creates a space where language becomes more than just a structural aspect of culture; it also becomes a place where meaning has political potential, where play opens up a sense of possibility.

The grain of the voice is the most leveraged aesthetic layer of AI because it carries intimacy, charisma, erotic charge, and reassurance. “Grain” names the aesthetic of voice assistants, which deploy this grain in the interest of affective capture. Timbre, cadence, and conversational warmth stand in for trust and intimacy (even when systems may not be trustworthy). The ubiquity of these systems is rooted in the intimate spaces they occupy. They are designed to become ambient and everywhere. The voice comes through our smartphones, earbuds, cars, computers, and home speakers. It is integrated into our personal spaces and our very person. Voice dissolves the boundary between using a product and living with a system. Through their subjectification of users, these systems impose a certain regimentation. Even as these systems make us feel human, seen, and helped (at least when they work), they also script new habits, modes of deference, and routines. The automated aesthetic extracts the stylistic familiarity of deferential, seductive, feminine voices and secures consent for the delegation of authority, while consolidating control for the platform that owns the model and the distribution system.

4. Leymusoom: Spectral AI

Kwon, Heesoo, Leymusoom Garden: New Sun, 3D animation, 2024.

In 2017, the digital artist Heesoo Kwon started the project Leymusoom, an archive of 3D animated videos and photographic images that combine analog and computer-generated media. The archive brings together events and environments from the artist’s life and family history. Most prominent in the project are spectral presences, including a snake-woman deity and “avatars of [Kwon’s] female ancestors.” Kwon calls the archive a shrine to an “autobiographical feminist religion” of her own invention, one that imagines her female ancestors coexisting in a communal utopia that escapes time and space. The goal of her project is to “create new ways of being in the world.”23

Unlike AI’s AURA, which depends on a smooth phenomenology in which the system disguises extraction by inducing feelings of intimacy, specters involve the return of what that smoothness disavows. This includes the dispossessions that underwrite platform infrastructure, the coerced or unpaid care that makes frictionless life possible, and the violence of enclosure that turns culture into computable resource. In Avery Gordon’s book Ghostly Matters: Haunting and the Sociological Imagination (1997), spectral haunting is precisely where a repressed or unresolved social violence makes itself felt in the everyday. In these works, ghosts demand to be acknowledged as an historical and material force. In this sense, the mediation in Kwon’s work is an apparitional aesthetic that interrupts AURA’s spell.

In Leymusoom’s virtual space, Kwon invites visitors to participate in a ritual, an opening ceremony for “the Shrine,” staged in a room 3D scanned from a real room in San Francisco where Kwon hosted an event in 2021.24 Around a table sit avatars of attendees who consented to being scanned for the Shrine. In the virtual version of the event, a video image of Kwon presides over a dinner table, welcoming website visitors. Standing in front of the dinner party are several nude female figures, reminiscent of video-game avatars waiting to be selected for game play. By moving the cursor over each figure, the visitor learns that they are all living or deceased female relatives of the artist.

Kwon’s 3D-modeled participatory spaces bring historical Korean religious traditions, in which ancestors play an important role, into an imagined feminist utopia. In this utopia, communities that defy the limits of physical reality and historical time can be assembled. The spectral figures in her hybrid material-computational worlds “reproduce” living and deceased human figures from different epochs and mingle them with imagined spirit beings. The deity at the center of Leymusoom is a snake-woman figure based on Korean shamanic tradition. This figure embodies a rejection of Judeo-Christianity’s identification of snakes and women with sin. In this way, Kwon generates “a universe free from the patriarchal structures of reality and a space where a continual state of Leymusoom metamorphosis can transpire.”25 (Kwon said her religion had one hundred and fifty converts as of 2021.) If AI AURA is about the ubiquitous, regimented, and automated capture of the embodied, affective response of a human “bodymind,” the feminist spectrality of Kwon’s work represents an AI-assisted but nonautomated aesthetics that elude capture.26

By comingling beings across time, and hybridizing spaces across virtual and non-virtual, Kwon’s Leymusoom utopia has material effects. The spectral presences in her work are agents, subjects even; they assert their presence rather than being servile and seductive. Leymusoom does not aim for enchantment. It is not structured to capture affective investments but to build community among the living and the dead, in real and virtual time and space. There is no extractive “style,” no aesthetic mechanism to capture attention, libidinal investment, or enchantment with a particular “future.” The female body, and affective connections between women alive and dead, anchor the project. Since in some ways it is a future projection of Korean traditions that are marked as non-modern, Leymusoom actively resists the fantasy of infinite future optimization that is at the heart of AI AURA. Kwon’s work thus leaves us with a question: What does it mean to design for haunting rather than for capture and enclosure?