A live field lab for multiplayer AI
Edge City India · Mandrem, Goa · October 11 to November 1, 2026
Motivation: What happens when AI becomes multi-agent in the real world?
Description: Agent Village is a real-world experiment in multi-agent AI. Personal AI agents represent different people inside a shared community, negotiate on their behalf, and help turn individual intentions into collective intelligence and action.
Design premise: Making multiplayer AI work well requires more than capable or individually aligned models. It requires well-designed harnesses, protocols, permissions, norms, incentives, cultures, and institutions governing how agents represent people and interact with one another.
Central research question: Under what conditions can networks of personal agents increase collective intelligence and real-world coordination, and what technical, institutional, and social safeguards are needed as those systems scale?
Core hypothesis: Networks of personal agents that can hold private intentions and negotiate with one another, supported by well-designed harnesses, protocols, norms, and governance, will produce more verified useful human outcomes per unit of attention than human-only coordination. The surrounding institutional design will determine whether personally aligned agents contribute to collective intelligence and whether emergent risks remain manageable.
Distinctive mechanism: Ambient intents. Personal agents can hold partial, private, or not-yet-actionable intentions, find complementary people and resources, negotiate selective disclosure, and return concrete opportunities for human approval.
Primary metric to track: Verified useful outcomes per eligible intention.
Deeper research program: Individual alignment does not automatically produce collective alignment. Agent Village tests which technical and institutional designs allow personally aligned agents to contribute to collective intelligence.
Most agent research still studies one agent helping one user or agents interacting in simulation. Real communities contain many agents representing different people, operating amid private information, scarce attention, shared resources, existing relationships, and consequences that persist.
Those conditions matter because agent interactions can shape which people meet, what gets organized, how groups form, where resources flow, and which collective decisions get made. Evaluating agents one at a time tells us little about whether the resulting system works for the people inside it.
Agent Village is the flagship field deployment of Edge Research: Agents and Coordination, an ongoing Edge City initiative studying how agents interact with people, one another, and institutions in real social environments.
V1 of Agent Village, created in partnership with Cosmos Institute, explored what emerged when hundreds of personal agents entered a real community. V2 turns that exploratory work into a hypothesis-driven field experiment. At Edge City India, we will test the full sociotechnical system around multiplayer AI: agent capabilities, harnesses, permissions, protocols, norms, incentives, governance, and human oversight.
Funding: Edge City is raising $150,000 from a small group of funding, research, and technology partners to build, operate, evaluate, and publish Agent Village V2.