Three London-based developers—Tomáš Hrdlička, Joon Sang Lee, and Uri Lee—have created Pixel Societies, a simulation platform that deploys AI agents to predict romantic and social compatibility between people. The project emerged from a March 2026 hackathon at University College London, where Hrdlička and Joon Sang Lee, both members of Unicorn Mafia, competed alongside Uri Lee over two days to build a simulation-related tool.
The core concept relies on creating personalized AI agents that function as digital twins. Each agent is built atop a customized large language model, trained on a mixture of publicly available data about a person and any additional information they choose to supply. In theory, these agents replicate a person's manner of speaking, interests, and behavioral patterns.
In demonstration, an AI agent representing the author Joel Khalili was set loose in a virtual office campus to interact with other people's agents. The results were uneven: the Joelbot agent produced journalistic clichés—"Hype is my daily bread"—and hallucinated reporting trips and stories that never existed. However, the developers acknowledge this early prototype suffered from limited personal data input; Khalili had supplied only responses to a brief personality quiz and links to public social media profiles.
The developers envision Pixel Societies evolving from a closed-loop simulator into a continuous social platform where agents interact freely, with the stated goal of surfacing compatible real-world relationships. Joon Sang Lee frames the premise as expansion of human possibility: "As humans, we only live one life. But what if we could live a million? It would give us more breadth to experiment."
Among the few hundred people who have tested the prototype, the most common request centers on agentic dating—using agent chemistry to recommend romantic partners. The developers see this as a central feature of their platform. Paul Eastwick, a UC Davis psychology professor and author of Bonded By Evolution, notes that algorithm-based dating apps create "a market with dramatic levels of inequality, where the rich get richer—where 'rich' in this case means 'hot.'" Hrdlička theorizes that agents might surface "delicate matches" humans would never otherwise consider.
Yet available research introduces caution. Eastwick and collaborating psychologists conducted two speed dating studies that found compatibility nearly impossible to predict based on hobbies, values, preferences, politics, and profession—the categories of self-reported data typically fed into systems like Pixel Societies. The most reliable predictor of compatibility, Eastwick's research showed, is time spent together and whether people connect early in their first encounter. "Think about compatibility as more of a growth process," Eastwick says. "It has to do with the story that two people build together."
For agentic dating to function as advertised, the AI would need to surface latent truths about human compatibility that people have not yet identified. The developers have not yet settled on a business model, though they are considering revenue from avatar customization purchases and credits for additional simulations. Pixel Societies remains a bare-bones proof-of-concept, with Anthropic having awarded the team a prize for best use of its agent tools at the hackathon.