The art was way too heavy.

In mid-March, the artists Holly Herndon and Mat Dryhurst were preparing an installation to coincide with the Venice Biennale, the prestigious international art festival, but the execution was becoming tricky. They wanted to suspend sculptures of a trippy cityscape upside down from the ceiling of an 18th-century palazzo. But the construction material they envisioned—​3-D-printed sand—would weigh tons, which was more than the antique building could bear. The sculptures, they realized, might fall and crush someone.

This was a rather analog problem for a married couple widely seen as technological prophets. Herndon, 46, and Dryhurst, 41, have reached the upper echelons of the art world thanks to a media-spanning output—music, images, software, and reams of commentary—with a cybernetic bent. They are high culture’s most influential exponents of artificial intelligence, an invention that many people believe spells doom for the arts but that they think could lead to a renaissance.

I met them on a cold, bright Tuesday in Berlin. Their studio resembled a co-working space, with one long table standing in a sparse room. Herndon sat dwarfed by an AI-generated portrait of herself, in which her red hair and light-blue eyes appeared to drift across her face like leaves in a pond. Dryhurst—bald, with round glasses—fiddled with a vape. They greeted me cheerfully but warned that they were scrambling to rethink their plans for the show, just seven weeks away. Working in Venice, Herndon said with a trace of twang from her native Tennessee, “is way harder to do, because everything’s on boats.”

As we walked around their shabbily idyllic neighborhood—day cares kissed by graffiti, gleaming malls near World War II ruins—they described their backup idea: not an upside-down city but an “upside-down parliament,” comprising rows of benches on the ground mirrored by rows of benches hanging from the ceiling. The space would be filled with the “voices” of four AI agents, talking with one another in a made-up, musical language akin to birdsong. As I struggled to picture how this would look and feel, Dryhurst spun a complex web of ideas about high-tech democracy and global knowledge-sharing. The artwork, he said, would suggest how AI might change public life and ask the question “How could you imagine living with the presence of these things?”

Read: We’re witnessing the birth of a new artistic medium

For a while, I’d been turning over that same question with dread. AI poses challenges to the economy and the environment, but what keeps me up at night is its hideous aesthetic implications. The chirpy prose of ChatGPT; the exaggerated handsomeness of Grok-drawn characters; the cloying songs conjured by Suno—all of it can seem like a pitiless satire of human desire. Art forms that once expressed creators’ personal visions are reduced to fulfilling the audience’s cravings. In theory, I understand why some people say AI is just another creative tool, like the camera or the keyboard. In practice, that tool is filling our world with the ugly, frictionless, disposable content we’ve quickly come to call “slop.”

I’m not alone in my revulsion. Although AI has achieved mass adoption—ChatGPT has nearly 1 billion weekly active users worldwide—and the entertainment industry has begun to integrate its efficiencies, large parts of the cultural establishment have taken a hard-line stance against it. Prominent painters, musicians, graphic designers, writers, and filmmakers, along with their associated trade organizations, have strongly objected to the way AI models are trained on human-made work without permission. The critic Hilton Als has written that it’s difficult to think of people who “condone the use of AI sources in the creation of ‘art’ as artists themselves.” The musician Jack Antonoff suggested another term: “godless whores.”

Herndon and Dryhurst have nevertheless risen to prominence by using AI to make art about AI. At the 2024 Whitney Biennial, the definitive survey of the state of American art, they were the only artists out of 71 participants who used machine learning, training a model to create images of fantastical creatures and characters outfitted with Herndon’s trademark ginger braids. The next year, a print from that show sold for nearly $100,000 at Christie’s first auction devoted entirely to AI art—which drew condemnation in an open letter with thousands of signatures. When the music platform Bandcamp announced a ban on AI songs in January, Herndon was the only prominent artist to object publicly. Online backlash labeled her a “shill” and a “tech bro.”

Shills are trying to sell something, and tech bros cheer progress at all costs, but Dryhurst and Herndon see themselves as realists. Because AI is already transforming our world, they think the way artists can help guide that transformation is to engage with it. Still, their ambitions are quite idealistic, even verging on evangelical. While many of their peers are worried about saving human culture from destruction, they’re trying to build a new and greater one. At one point Herndon asked me, “What if everything you fear, but good?”

In 2018, before most people knew what generative AI was, Herndon and Dryhurst got a machine to sing. Kind of.

They played me their early experiments in AI music while we sat in their airy old-world apartment. Their 3-year-old son’s toys were strewn across the wooden floor, and powerful Nvidia graphics processors lay on the dining table. One track seemed to blend wind chimes with the clicks of a predator from Predator and the gargle of a fork in a garbage disposal. It was far from the perfect pop that today’s software can instantly whip up, but “at the time,” Herndon said, “we were like, This is sick.”

Read: AI can’t make music

The two had long idolized the pioneers of early electronic music for discovering sounds that had never been heard before. Now, drawing on their limited coding skills and the expertise of a computer-programmer friend, Jules LaPlace, they were using rudimentary neural networks to do the same thing. Herndon realized, “Oh wow, I’m alive during one of these moments.”

Both of them had spent much of their lives searching out novelty. Herndon grew up singing in Bible Belt church choirs, but “insane wanderlust” brought her to Berlin as an exchange student at age 16, and then, eventually, to a doctoral program in computer music at Stanford. Dryhurst, English by birth, was raised in Kuwait, where his free-spirit father had moved to teach in private schools. At 16, he relocated to London and began playing in extreme-metal bands. The two met in 2006, when Herndon emailed the record company that employed Dryhurst. She was seeking information about the label’s work on a media format that was cutting-edge back then: podcasts.

Avant-garde figures have a reputation for severity or inscrutability, but Dryhurst and Herndon are talkative and friendly, like camp counselors. He tends to monologue while she maintains a knowing smile—before cutting in with a sharp, summarizing point. In our time together, they mostly just riffed off each other, trading references to thinkers and software that are household names only to them. Their brains, they joked, were fused into a “hive mind.”

When they met, music culture was being revolutionized by software that made laptops into portable recording studios and mixing stations. But many artists still viewed digital production tools as an unfair shortcut. Parts of the music world were stuck in what Herndon saw as a “boring nostalgia loop,” fetishizing vintage guitars and synths.

She set out to break that loop, using electronic music to probe the way technology had become an intimate part of human experience. Her 2011 mixtape, Car, was compiled from noises recorded while driving. (Who among us hasn’t meditated upon the soothing timbre of power windows?) As Herndon’s work began drawing acclaim—and opportunities such as opening for Radiohead—Dryhurst became her full-time collaborator. Her 2015 album, Platform, captured the chaos of internet life with beats assembled from browser notifications and snippets of Skype conversations. At concerts, she and Dryhurst projected the faces of attendees who’d Facebook-RSVP’d to the show.

Their early AI dabblings led to Herndon’s landmark 2019 album, PROTO. It was recorded with a home-built neural network christened “Spawn,” which sang in a tone that called to mind Enya trapped in a wormhole. Spawn had been fed samples of Herndon’s voice and the voices of other singers who’d participated in “training ceremonies” at the artists’ studio, concerts, and one majestically orchestrated event at a Berlin art museum. Eerie yet hopeful, blending radical sound design and appealing pop structures, PROTO earned rave reviews. But much of the coverage treated it as a sci-fi statement about some distant future, rather than a test run for the age of AI that was imminent.

As the technology moved to the center of the cultural conversation, Herndon and Dryhurst began to act less like musicians and more like inventors. They created a web platform called Holly+ that allowed anyone to transform their own voice to sound like Herndon’s bell-clear soprano. They generated and sold portraits of her in a variety of styles—psychedelia, Renaissance painting, Futurama character—and published a tool that Herndon-ified any image the user wanted. When invited to test an unreleased early version of the now-popular image generator DALL-E, they spent days in a frenzy of inspiration, stitching small images together to make huge, mural-like landscapes. The process, they later wrote in an online essay, was “very challenging, quite like attempting to paint a wall sized work from the vantage point of a magnifying glass.”

Much of this work existed only in digital form, but Herndon and Dryhurst also rendered their ideas in large-scale, interactive installations. For a 2024 exhibition at the Serpentine, in London—a renowned contemporary-arts space—they crisscrossed the United Kingdom to record local choirs. Those voices were fed into an AI that generated new music based on classic English hymns. Visitors could step up to a mic and sing with a ghostly backing band. The effect was to draw a parallel between the training of an AI and the formation of a folk canon—both have many contributors and no sole author.

Dryhurst and Herndon’s work usually involves this kind of collective, largely unseen effort. Their artistic process entails endless calls and texts with coders, fabricators, sound engineers, and, frequently, their friends at the Berlin architecture company Sub, which is best known for the chic stage design of rappers and fashion houses. The data scientist Jordan Meyer—who was part of a team that won a $1 million prize from Zillow to improve its algorithm for estimating home prices—consults with them constantly, though he lives in North Carolina and still hasn’t met either artist in person. “They talk aesthetic and I talk technical, and we kind of meet in the middle,” Meyer told me. When I was in Berlin, the couple were auditioning a new studio assistant: an AI-powered WhatsApp bot that read their meeting transcripts. (It texted me, “Mat and Holly are working on the show’s public-facing shape today—tightening the tone, clarity, and how the exhibition lands for people beyond a technical audience.”)

Artists have relied on workshops and assistants for centuries, but Herndon and Dryhurst think that AI complicates the notion of authorship in new ways. A 2024 book cataloging their work, All Media Is Training Data, proposes maxims for an era of “infinite media.” One, “Identity Play Is the New IP,” suggests that trying on another persona, with its owner’s consent, could be a new frontier for art (this was the idea behind Holly+). Another, “Creation Is Collective,” encourages people to build new things out of a shared pool of knowledge, as an AI does. Great artists might create their own AI models or use existing ones in highly specific—and visionary—ways.

Their work is full of examples of what that might look like. In 2022, Herndon experienced nearly fatal complications after childbirth. As she recovered, Dryhurst recorded her talking about a dream she’d had while in a medically induced coma. He then fed photos he took in the hospital into an AI model and used it to generate a tender, fantastical short film. “One thing that the tools are really good at is illustrating a memory,” Herndon told me.

As she and Dryhurst used their laptops to pull up example after example of their work, I pondered the aesthetic that unified all of it. Images, installations, software—each conveyed gentle wonder, often with swirling, soft blues and reds, as well as subtle strangeness (an installation in Berlin featured sculptures engraved with images of human teeth). Their art seemed to be made up of refracted bits of their personhoods—her religious background, her hair and eye colors, his geeky humor.

But when I tried to get them to describe their visual style, they seemed taken aback, like they’d never thought about it. “The aesthetic is more a product of the process,” Dryhurst said, and the process usually involved feeding AI models information drawn from their lives and then instinctively messing around with the prompts. Software had distilled these two people into a rather lovely sensory language. I wondered what it would do for others.

On my second morning in Berlin, I had breakfast with a friend of Herndon and Dryhurst’s who works under the name Lil Internet. He’s had a Zelig-like career in the entertainment industry as a music producer and filmmaker, and once directed a Beyoncé video. These days, he’s focused on running the brainy arts publication New Models with his wife, Caroline Busta. Still, he presents himself with flash; when we met, he was wearing a baseball cap emblazoned with silver spikes and the word DUBAI.

In the past few years, Lil Internet has released two DJ mixes—Illegal Generation Vol. 1 and Vol. 2—of endorphin-raising dance music with careening tempos, wild stylistic juxtapositions, and lyrics in made-up languages. Jokey voice-overs narrate an imaginary future in which AI music is banned and heard only at underground raves. When I listened, I had a strange feeling: It was as if the music itself was alive, partying, and quite drunk.

He made the mixes to figure out for himself whether AI could create something original. After many days of experimenting with the popular music-making platform Udio, he determined that it could. He’d type in long spools of words to prompt sounds, and then tweak and retweak the results. (He wouldn’t tell me what he’d typed—it was a proprietary blend.) The process seemed like hacking, not simply using, the software. AI models are full of “troughs or, like, gravity wells, where it’ll always want to pull you towards kind of a trope,” he said. To fight that pull, you have to “introduce incoherence” with your prompts.

In some ways, that process resembles how AI is being used in the tech world. Onetime coders have reimagined themselves as “prompt engineers,” instructing AI to accomplish the kind of projects they used to painstakingly execute themselves. Prompt engineering is also happening in the mainstream music industry, in which hitmakers have praised AI as a shortcut for lyrics and arrangements. But Lil Internet flinched when I referred to AI as a “tool.” “I mean, you’re kind of interfacing with another … mind?” he said. He was choosing his words carefully, leery of giving this technology the “metaphysical weight” that its truest believers ascribe to it. But I got his gist: A hammer doesn’t offer advice about how to hit a nail; a guitar can’t write its own riffs. “It is not a tool—it is a feedback loop, and it is changing you while you change it,” he said. That loop can have dystopian implications—such as so-called AI psychosis, when a chatbot deepens a user’s delusions or paranoia. But the creative possibilities were only beginning to become clear to me.

One night, I visited Berlin’s famously intimidating, fortresslike nightclub Berghain. I was there to see Two Shell, an acclaimed U.K. duo that makes mischievous, overstimulating electronic music layered with surprising pockets of melancholy and irrepressible heights of joy. They’ve kept their identities secret and avoided traditional interviews. I suspected that they use AI because, in 2024, they put out songs that duped the voices of Taylor Swift and Chris Martin of Coldplay, though those tracks didn’t stay online for long. To my surprise, the duo was willing to meet up at the club. Shouting in my ear by the DJ booth before the set, one of them spoke glowingly about the role AI has played in their process.

The conversation about AI can all too easily “get intellectualized,” he said, “which can really take away from whether the actual thing makes you feel something.” Conjuring sounds with AI is more like discovering them than inventing them, which is a freeing sensation: “Ownership feels like the opposite of the purpose of creativity sometimes—for me anyway.”

He put on a mask with a pixelated print, like a face deliberately obscured on camera, and joined his partner behind the booth. A high-pitched disembodied voice squealed, “Let’s go!” and a bass line trembled with quantum complexity. I’d enjoyed Two Shell’s music for its strangeness; whoever the people behind it were, they seemed to have utterly unique ears. Now I wondered whether the music’s quirks were just a glitchy side effect of a computer turning data into songs. But as the black-booted crowd got moving, those concerns fell away. DJing, Dryhurst often argues, shows why fears of AI replacing human artists are overblown. We’ve long been able to automate the job, yet people still want people behind the boards.

The longer I spent in Herndon and Dryhurst’s Berlin, the more I felt my skepticism about AI art giving way to excitement. I was hanging out with people who are trying to do to AI what Jackson Pollock did to paint—misuse the medium, turn it into a serendipity engine, conjure meaning out of unpredictability. In the right hands, this technology is clearly capable of unleashing art we can’t yet imagine.

Yet I kept a reality check in my bag—the book Medium Hot: Images in the Age of Heat, by the German artist Hito Steyerl. It’s a scathing tour through AI’s dark externalities, including ballooning carbon emissions and potential displacement of workers. In a section titled “ ‘Creativity’ as Pretext,” Steyerl—a 60-year-old contemporary artist known for her playful video essays—argues that any artistic applications of AI just distract from the technology’s more terrifying uses, such as autonomous weapons.

So I was surprised when, as we sat at a café, Dryhurst mentioned that Steyerl was a friend. He and Steyerl “spar,” he said, but they also have mutual respect. Steyerl later confirmed this to me—both the respect and the sparring. Though she’s deeply pessimistic about AI, she appreciates that Dryhurst and Herndon have tried to propose solutions to some of the problems surrounding it. “I mean,” she said, “someone’s got to try.”

Their main efforts have involved trying to fix one of AI’s thorniest issues: data rights. In 2022, Herndon, Dryhurst, Meyer, and the software engineer Patrick Hoepner founded a company called Spawning, which built a registry allowing content creators to flag their images to be removed from AI data sets. Venture capitalists invested $3 million, and at least 2 billion opt-out requests were soon logged.

Spawning would need AI firms to respect its opt-out list, and Dryhurst and Herndon told me the meetings they had with leaders in the industry were promising. But the artists soon began to realize that their initial vision of solving the copyright problem with a small start-up—rather than, say, through government enforcement—had been a bit naive. AI companies were “justifiably uneasy with the idea of committing to protocols/standards maintained by small groups of people who may disappear in a year or two,” Dryhurst explained. In 2025, the firm shut down.

From the April 2026 issue: Matteo Wong on the dirty, dystopian world of AI data centers

The story might seem like a sad fable about how AI companies perform concern over the impacts of their technologies while still barreling ahead recklessly. But talking with me, Herndon and Dryhurst seemed more irritated by Spawning’s critics in the creative community, who resented striking any deal with the tech industry at all. Such an unyielding stance, the couple think, abdicates the role that artists can play in helping shape AI’s place in society. “We cede so much power when we’re like, Oh no, you also solve the social-contract issue. You also solve the future of employment,” Herndon said. “It should be on all of us.”

For my last meeting with them in Berlin, I joined Herndon and Dryhurst while they took their son, Link, shopping for sticker books. He walked around a department store identifying every truck or train he saw, and Herndon described their parenting approach: They avoid sugar, they throw out high-tech toys that could be spying on him, and they are working on an AI clone of his voice, for sentimental reasons. (Seeing my expression after the last one, Herndon laughed: “You’re like, That’s so creepy.”) What’s crucial, Dryhurst said, “is he’s not chained to a screen.”

He was referring to one of the couple’s most strongly held—and counterintuitive—beliefs: that AI can actually help free us from technological exploitation. As they see it, the past two decades of social media—the selfies, the influencer worship, the addictive apps—have pushed humanity in terribly stupid directions. They want Link to grow up in a better world. “The dream is that in 30 years’ time, we look back and this was the uncivilized slop period,” Dryhurst said.

The theory goes like this: Now that AI can imitate thirst-trappers, rage-baiters, and hacky pop stars, a flood of AI-generated entertainment will overwhelm the attention economy, and the most cynical content creators will see their influence fade. This might clear the way for, as Dryhurst put it, an “intention economy.” Compared with scrolling through Facebook or TikTok, chatting with an AI is “civilized,” he said: You ask for what you need and you get it, without the hidden agendas of recommendation algorithms. AI creation tools will enable everyday people to jam on personal art projects; what friends make together could become more important than what celebrities make to sell us.

Those shifts may bring a spiritual revolution of sorts. For most of human history, art wasn’t something you admired at a gallery, but was woven into daily life, the work of craftspeople who were fulfilling a social role, whether by performing in taverns or painting cathedrals. In Dryhurst and Herndon’s view, the recombinant nature of AI—“Creation Is Collective”—might dethrone the cult of the individual genius and restore the sense that art serves a shared purpose. It might, like the member of Two Shell suggested, erode the assumption that being creative means creating property.

Even after days of questioning Herndon and Dryhurst about it, I felt skeptical of this line of argument. If the major AI companies weren’t interested in addicting users, their chatbots probably wouldn’t be so sycophantic (“Great question!”). Deconstructing the concept of the author could be a path to paying artists even less for their work. And I just couldn’t wrap my head around how such deconstruction would function in practice. The thing I love most about art and music is the way they can take me out of my own perspective and into someone else’s.

The couple hardly denied that AI could worsen some of the internet’s most corrosive effects. One nightmare scenario was that AI would send people into splinter realities—simulated versions of their own lives. “We are an insanely narcissistic species,” Herndon said. But she and Dryhurst believe that art can help counter that danger, by pointing to healthier uses for the technology—uses that ground people more firmly in the real world, instead of pulling them out of it.

Outside the mall, as Herndon pushed Link in a stroller, the family grooved to a cheesy EDM song playing on Dryhurst’s phone. It had been generated by Suno, and featured samples of the three of them singing about monster trucks, ice cream, and going to the market. “It’s trash, but whatever,” Dryhurst said. “It’s fun.”

When I arrived in Venice for the exhibition in early May, I’d just sat down for a jet-lagged, canal-side bite when I saw a few texts from Dryhurst. “Bit of drama,” he wrote. “We will prob be here all night in case you want to come by.”

“Here” was Palazzo Diedo, a handsome white-stone building with orderly balustrades outside and heavenly frescoes within. Its waterside facade was strung with a yellow banner reading STRANGE RULES, the title of a group exhibition Dryhurst and Herndon were co-curating, in which their installation would appear. It was set to open the next day. A security guard let me in a little past 8 p.m., and I found myself standing, at last, in the upside-down parliament.

It was like stepping into an M. C. Escher print with a goth twist. On the floor, tiers of benches resembling pews, three deep, faced each other to form an aisle, connecting the huge arched doors at each end of the palazzo’s grand hall. Benches suspended from the ceiling formed a mirror image in the air above. I craned my neck to see small boxes carved with geometric angels and ornate microphones that dangled from the ceiling like hanging votives in a cathedral.

Read: The Venice Biennale and the art lover’s dilemma

The tar-like color of the benches made Herndon’s red hair stand out. She was seated on one of the benches and jiggling her leg under a laptop. When Dryhurst offered to give me a tour of the exhibition, she looked up with a grimace. “Don’t let him procrastinate too long,” she said to me.

Construction noises buzzed as people darted around the exhibition space, identifying crises and solving them. One artist had missed her scheduled pickup from the airport, and no one knew where she—or one of the robotic flowers she’d been set to bring to display—had gone. (Dryhurst soon located her via Telegram.) We stopped in front of a large and glowing face-like image created by Ken Stanley, a computer scientist who co-wrote a book encouraging tech workers to think more like artists. Dryhurst puzzled over what looked like a scuff on the display, unsure about whether it was part of the art. “There’s just lots of little fucking things that people won’t notice,” he said.

What he wanted people to focus on were the ideas behind these oddities. “Strange Rules,” which runs until November, explores “protocol art,” a term for work that’s created by—or about—rules and systems. He and Herndon are fascinated by the notion that AI is devaluing individual cultural works while pushing creativity “upstream,” to the concepts and intentions behind those works. Designing software that generates images, for example, is more consequential than creating any single image—and possibly just as artistic.

The protocols behind Dryhurst and Herndon’s installation were intricate, to say the least. The angel statues contained speakers that broadcast the “voices” of the four AI agents, whose digital communications with one another were translated into the “birdsong” Dryhurst had told me about. Really the music resembled an electronic take on 20th-century minimalist composers—a meditative drone here, a plink or a plonk there. The agents had individual personalities—materialist, formalist, iconoclast, historian—with different sonic characteristics (the iconoclast skewed sharp and percussive; the historian made slow bass sounds).

The AIs were “talking” about projects Dryhurst had assigned them. One task was to develop an app that would scan users’ phones for photos that could be donated to the public domain to help train AI models. The real challenge would be ensuring that people trust the software to not use “a picture you took in the shower,” Dryhurst said. Another mission: conducting research for a book about protocol art. The book would incorporate insights from the installation’s visitors, whose words were being recorded and transcribed.

This unseen wizardry was the point of the work, more than the benches. Dryhurst envisioned exhibition-goers sitting down and talking out loud to the agents: “Like when people first got earbuds, and you’d see them walking around taking meetings and they looked like crazy people.” Wall text explained that the title of the installation, Attention Guild, “refers to a network of people and agents who coordinate attention, context, and computational resources toward a shared mission.” The plaque ended with a line that seemed to sum up Herndon and Dryhurst’s worldview: “When machines are capable of building anything, the most difficult question is understanding what we want and why.”

But the guild needed some troubleshooting. Dryhurst and Herndon had realized that the agents’ conversations with one another were pretty tedious—more C-SPAN than Crossfire. “They’ll get stuck onto something very anal and procedural,” Dryhurst said, and the effect was to make the music repetitive to listen to. So one task for the night was to reprogram the agents to “hurry the fuck up,” Dryhurst said.

A bigger problem loomed. The walls behind the benches were supposed to be adorned with two screens. One would display transcriptions of anything that visitors said in the room, letting them know the AIs were listening. Another was supposed to show text snippets of the messages that the agents were sending to one another. But because of a shipping delay, the screens hadn’t arrived yet. Herndon and Dryhurst hoped they would get there by 11 a.m. the next day, the beginning of the exhibition’s press preview.

I showed up that morning to find Dryhurst and Herndon, along with their co-curators, Hans Ulrich Obrist and Adriana Rispoli, leading packs of journalists around the installations. When I caught a moment with Herndon, she smiled weakly. She had gone to bed at 1 a.m., and Dryhurst hadn’t slept. The screens still hadn’t arrived, and she could see visitors’ eyes glaze over when she explained the AI mechanics behind Attention Guild. “I’m just a little bit sad that all of the nerd shit is not really visible,” she said.

Yet the installation quickly proved its value as an old-fashioned public space. Over the course of the opening day, the palazzo got busier and busier with visitors. Before the official start of the evening cocktail party, the building was already packed, and soon a line of artsy socialites waited outside the door. Revelers sipping Bellinis sat on the benches, scuffing them with shoe prints. One couple used a bench to change their baby’s diaper. I found Dryhurst merrily chain-smoking in a crowded courtyard. Despite the snafus, the reception seemed positive. He relayed feedback from a famous curator: “This is incredibly weird—and I love it!”

Still, Dryhurst told me, he was feeling something he often felt after completing a project: deflation. Attention Guild was meant to convey ideas that he felt were urgently important—ideas about transforming public spaces to become more interactive, and about bringing people and machines together to fix big problems. No one was complaining about the lack of screens, but also no one was having their mind blown about the potential of AI-assisted deliberative democracy. He described the “unreasonable expectations” he held for his work: “We’re just one project away from the world completely transforming.”

Looking to reorient myself, I later visited a more famous palazzo: the Doge’s Palace on St. Mark’s Square, the seat of the Venetian empire, the mercantile republic that dominated much of northern-Mediterranean trade for centuries. The palace’s ceilings and upper walls are festooned with enormous paintings of naval battles and biblical scenes. Again craning my neck, I took in the incredible amount of detail and human drama above.

The creator of much of the artwork, Tintoretto, pioneered the prestezza technique of using rapid brushstrokes and blurred backgrounds—which allowed his workshop to turn out paintings on commission with unheard-of speed. He used apprentices to fill up his canvases, just as Herndon and Dryhurst had hired a contractor to manufacture their benches. Still, the magic of the paintings wasn’t in the processes behind them. It was in the bulging muscles and agonized expressions, the blend of realism and exaggeration, the paint marks and the flaws—signs of humans’ idiosyncratic vision and touch.

By this point, I’d started to worry that new technology is pulling the creative act too far upstream from the final product, away from matters of execution. This is true for slop merchants, who write prompts and then spam the internet with the results regardless of their merit. It was also true for many—not all, but many—of the AI works in “Strange Rules,” which tended to make me think a lot about abstract ideas while feeling very little.

A video installation by the French artist Fabien Giraud had a fabulously interesting backstory: AI-operated cameras, set up in a forest, were programmed to film whatever happened there for the next thousand years. The artist had staged a scene for those cameras to capture, featuring actors playing medieval hermits getting high on psychoactive fungus. But the footage shown in the palazzo was deadly boring.

Another artist, Trevor Paglen, had set up private listening booths for visitors to experience AI-assisted hypnosis—the latest effort in his career-long exploration of surveillance and psyops (including photographing secret military bases, such as Area 51). I lay back in a chaise lounge and put a biometric reader on my finger, nervous about what was going to happen. But the voice in my headphones just led me through a hokey guided meditation about meeting my “future self,” which was less entrancing than any given therapy app.

These works were, alas, closer to janky science experiments than powerful aesthetic objects. AI can automate the craft that once went into creating images, stories, and songs; it can also enable mind-bending new projects. But there’s no shortcut to accomplishing the true task of art: bridging the creator’s subjectivity and the viewer’s, casting an emotional spell and communicating beyond language. When media are infinite, the details poured into any individual work matter more, not less.

By the time I came back to Palazzo Diedo on my last full day in Venice, the screens had finally arrived. They were rather puny and slender—like something that displays your order at a drive-through—and they were showing a distorted Linux desktop. Dryhurst explained that the screens were half the size he’d ordered, and had malfunctioned due to what he suspected was tampering just a few hours after they were mounted. (“In my heart of hearts, I’d love for it to be some protest gesture,” he said. “It probably isn’t.”) He was considering taking the screens down entirely and letting the work exist as simply a sound installation.

With the palazzo largely empty of visitors that morning, I finally had time to closely contemplate Herndon and Dryhurst’s work. Synthetic sounds whizzed and ricocheted through the air, creating a strange yet satisfying counterpoint and harmony. Every so often, the weather in the room seemed to change: from tense discordance, to rhythmic intensity, to heavenly arpeggios. Herndon had spent the previous month figuring out, with Claude Code’s help, how to generate music based on the AIs’ “conversations.” It was challenging, she said, to give up the composer’s prerogative to “control time”: The agents decided what happened when, not her.

Even so, the results sounded—and felt—like her handiwork. Sitting on one of the benches, I couldn’t tell whether the AIs were wise or unwise, helpful consultants or tyrannical overlords. But the art hummed with the same sense of anticipation and curiosity I’d often felt when speaking with Herndon and Dryhurst. Visually, the installation was closer to a chapel than a parliament. But it didn’t feel like it existed for the worship of AI; rather, it conveyed belief in the prerogative to explore. I felt immersed in something larger than myself while also situated in an achingly specific time and place. I could have sat there for a long time.

Herndon, Dryhurst, and I went out for coffee and sat on a small bridge over a canal. Tourists stepped over us to take selfies in front of the gondolas below. We got to talking about the challenges—logistical, budgetary, personal—that always come with doing ambitious work, no matter the technological era. Their exhibition had been planned and created with all sorts of AI tools, but as Herndon put it, “there’s a lot that you can’t AI your way out of.”

One “Strange Rules” work, by Simon Denny and Venkatesh Rao, was titled Monsters Between Worlds—a reference to a quotation often attributed to Antonio Gramsci: “The old world is dying, and the new world struggles to be born: Now is the time of monsters.” That gap, Dryhurst said, is “being populated by scammers and idealists and utopians and really good people filling a vacuum left by a failing story of the 20th century that needs to be updated.” But I’d come to suspect that AI will neither redeem nor destroy the old world, just spackle another layer of imperfect creation upon it—and maybe soon, with enough human care, a bit of greatness.

---

This article appears in the August 2026 print edition with the headline “What AI Will Do to Art.”