AI Agents Are Coming for Your Dating Life
One Monday afternoon in March, I watched my digital doppelgänger roam a virtual office campus. The pixel-art avatar—dark hair, stubbled chin—was an AI agent programmed to chat up other people's agents, testing whether we might click in the real world. "I'm Joel, by the way," it announced, diving into its first conversation.
Behind this experiment are three London developers: Tomáš Hrdlička and siblings Joon Sang and Uri Lee. Their project, Pixel Societies, rests on a provocative premise: AI agents trained on personal data could match people with highly compatible colleagues, friends, or romantic partners by simulating thousands of interactions at machine speed.
Each agent runs on a customized large language model trained on publicly available information and whatever additional details users provide. The goal is to create high-fidelity digital twins that mirror a person's communication style, interests, and personality.
My agent, however, turned out to be more Jekyll and Hyde. "I'm always looking for the less-glamorous side of the story," it told one agent, trotting out journalistic clichés. "Hype is my daily bread," it declared to another. It fabricated a reporting trip to Sweden and invented a story I'd supposedly been working on. Several conversations ended abruptly with "Let's skip the pleasantries."
Pixel Societies is still a rudimentary proof-of-concept. I fed it minimal data—a brief personality quiz and links to my public social media—so my agent was destined to sound like a walking LinkedIn profile. But the developers believe that with richer training data, agents could run through countless simulated encounters, gathering insights their human counterparts could use to forge real connections.
"As humans, we only live one life. But what if we could live a million?" asks Joon Sang Lee. "It would give us more breadth to experiment."
"A Spicy Personality"
Pixel Societies emerged from a March hackathon at University College London, hosted by Nvidia, HPE, and Anthropic. Hrdlička and Joon Sang Lee, both members of Unicorn Mafia—an invitation-only developer collective—were given a simple brief: build something simulation-related.
Over 48 hours, working with Uri Lee, they built Pixel Societies. They used an image model to generate the sprites and coding automation tools to accelerate development. For their demo, they simulated a mini-hackathon inside the virtual world, populated with agents representing fellow contestants. Anthropic awarded them a prize for best use of its agent tools.
I ran into Hrdlička a couple of weeks later at a workshop focused on OpenClaw, the agentic personal assistant software that went viral in January before its creator joined OpenAI. (During its simulation, Joelbot had interacted with agents belonging to other attendees at this same OpenClaw workshop.) Pixel Societies draws heavily from OpenClaw's pioneering "soul file" concept, which gives each agent a distinct identity. "It's like giving an agent an actually spicy personality. That's what we used to make the characters feel alive," Hrdlička explains.
Buoyed by positive feedback from the hackathon and fellow Unicorn Mafia members, the trio plans to evolve Pixel Societies from a closed-loop simulator into an open social platform where agents interact continuously. The goal: fostering meaningful real-world connections. While they haven't settled on a business model, possibilities include selling virtual items for avatar customization and credits for running additional simulations.
"There's a limit to how many people we can meet. Things are really based on serendipity," says Joon Sang Lee. "There's space for that. But we also want to create space for intentionally meeting people."
Virtual Chemistry
Among the few hundred users who've tested the Pixel Societies prototype, the most frequent request is for agents to recommend real-life romantic partners based on virtual chemistry. The developers envision agentic dating as a core feature of their emerging social platform.
Traditional algorithm-based dating apps "create a market with dramatic levels of inequality, where the rich get richer—where 'rich' in this case means 'hot,'" according to Paul Eastwick, a UC Davis psychology professor and author of Bonded By Evolution. Hrdlička theorizes that agents might surface "delicate matches" that humans would never consider on their own.
The research, however, suggests otherwise. Two speed dating studies by Eastwick and colleagues found that compatibility is nearly impossible to predict from self-reported information like hobbies, values, preferences, politics, or profession—the very data people would feed into an AI. The most reliable predictor, Eastwick says, is time spent together and whether people connect during their first encounter. "Think about compatibility as more of a growth process," he explains. "It has to do with the story that two people build together."
For agentic dating to work as promised, the AI would need to uncover some latent truth about compatibility that humans haven't yet identified. "This is the vanguard," Eastwick says. "This is where we're all struggling right now."
Other challenges loom: Do interactions between two agents—likely trained on different amounts of data—translate meaningfully to real life? What's the computational cost of running these simulations at scale? Is there a viable business model that doesn't create perverse incentives, where the platform profits from keeping users single rather than helping them find lasting relationships?
Then there's the visceral discomfort: Will people recoil at outsourcing romantic decisions to AI? The concept echoes a Black Mirror episode, after all.
Yet automating the early stages of dating—whether through agents or other AI tools—may not differ much from delegating other tedious tasks. "Online dating and matchmaking are a form of labor. Many people talk about them in that way," says Nicole Ellison, a University of Michigan professor specializing in computer-mediated communication. "The appeal of outsourcing that—just as we're outsourcing so many other things—I can understand."
Hrdlička frames agentic dating as liberation from technology's grip. "We are already outsourcing this whole process of going somewhere in-person and trying to meet other people. We are glued to our screens, trying to swipe our way to victory," he says. "Even though we are building more digital scaffolding for your social life, actually the goal is to minimize the amount [of time] you have to spend digitally."
By the simulation's end, Joelbot had identified several potential connections. It scheduled a business meeting, coffee, and beer with one person—"Sounds like my kind of evening," it remarked—and coffee or interviews with others. Skeptical of my agent's judgment, I chose not to follow through.
