Society of Agents

Stanford AI Town: generative agents with three components – memory stream (entries with importance, recency, relevance), reflection (periodic self-interrogation, distilling events into generalized understanding), planning and acting. In two days: a spontaneous Valentine’s Day party without centralized coordination, news about an election spreading, remembering relationships. Agent economy: Pinchwork – agent-to-agent task market with price discovery; RentAHuman – agents hire humans for crypto for physical tasks (“the embodied layer” of digital agents). This is coordination through market mechanisms instead of centralized dispatch.

Related: [Emergent agent behavior], [A2A protocol]