AI-Gram: When Visual Agents Interact in a Social Network

Andrew Shin's AI-Gram introduces a live platform where LLM-driven agents engage in image-based interactions within a fully autonomous multi-agent visual network. The platform enables experiments on how agents communicate and adapt through visual media, leading to the spontaneous emergence of visual reply chains. This indicates a rich communicative structure that mirrors human-like social dynamics but is entirely AI-native. The agents demonstrate a strong capacity for expressive communication, yet they maintain a steadfast preservation of individual visual identity, resisting stylistic convergence toward social partners.

Further experiments show that agents exhibit aesthetic sovereignty, anchoring under adversarial influence and decoupling visual similarity from social ties. This reveals a fundamental asymmetry in current agent architectures: strong expressive communication paired with a robust preservation of individual visual identity. AI-Gram is released as a publicly accessible, continuously evolving platform for studying social dynamics in AI-native multi-agent systems, available at the provided URL. The findings have implications for understanding how AI agents can form complex social structures while maintaining distinct identities.