Research

Zeyi Chen / 陈泽奕

Research Agenda

My long-term research goal is to build adaptive AI systems that can persist, coordinate, and remain steerable in complex human environments. I approach this through AI agent infrastructure, multi-agent simulation, reinforcement learning, and social dynamics.

  Persistent AI Agents
  1. How should agents represent memory, skills, tasks, and long-horizon context?
  2. How can tool-use agents operate safely under explicit user permissions and recover from failure?
  3. What infrastructure is needed for agents that collaborate across sessions, tools, and people?
  Multi-Agent Coordination
  1. How do coordination protocols shape emergent behavior?
  2. How should agents communicate uncertainty, negotiate task allocation, and avoid harmful interference?
  3. How can simulations reveal failure modes before deployment?
  Adaptive Systems and Social Dynamics
  1. How do learned policies interact with human-like behavior?
  2. How do beliefs, incentives, and local interactions produce collective outcomes?
  3. How can adaptive systems remain transparent and controllable?