Social Simulation / Agent Society

Definition

Use multiple agents to simulate a crowd, organization, community, or social system. The focus is long-term memory, planning, relationships, and emergent behavior.

Category: Simulation

When to use

User research, product validation, social behavior simulation, game NPCs, organizational modeling, information-propagation studies.

When not to use

Production task execution, strongly deterministic flows, anything that needs a coding agent to perform real work.

How to implement

1. Design world state: locations, time, events, objects, relationships.
2. Each agent has memory, profile, goals, and a daily plan.
3. Use an observe → reflect → plan → act loop.
4. Record agent interactions and world-state changes.
5. The goal is simulation credibility, not single-task success rate.

Minimal pseudocode

async function tick(world) {
  for (const agent of world.agents) {
    const obs = world.observe(agent);
    agent.memory.store(obs);
    const reflection = await agent.reflect();
    const plan = await agent.plan(reflection);
    await world.apply(await agent.act(plan));
  }
}

Recommended trace events

• simulation.tick.started
• agent.observed
• agent.reflected
• agent.acted
• world.state.updated

Common failure modes

• Treating simulation results as real predictions.
• Persona is convincing but behavior isn't verified.
• Long-term memory pollutes future runs.

Implementation checklist

• Trigger and exit conditions defined.
• Input/output schemas defined.
• Permission, budget, timeout, and retry policies defined.
• Trace events defined.
• Degradation or human-takeover strategies defined.

References

• Generative agents