llm-agent-economies-information-limits

name	llm-agent-economies-information-limits
description	Pre-registered experiment on small economies of frontier LLM agents (Claude Opus 4.8), testing information-theoretic capacity regions for wealth growth under market coupling and mean-field residual attractor dynamics. Activation: LLM agent economies, information-theoretic capacity, mean-field dynamics, attractor dynamics, market coupling, multi-agent simulation, wealth growth.
metadata	{"arxiv_id":"2607.06001","published":"2026-07-07","authors":"Cheng Qian","tags":["llm-agent-economies","information-theory","mean-field-dynamics","attractor-dynamics","market-coupling","multi-agent-simulation","wealth-growth"]}

Information Limits and Attractor Dynamics in Economies of Frontier LLM Agents

Overview

A pre-registered, two-part experiment on small economies of frontier language-model agents (Claude Opus 4.8), testing two quantitative predictions about coupled multi-agent systems: an information-theoretic capacity region for wealth growth under market coupling, and a mean-field residual-attractor dynamics model.

Key Innovations

Information-Theoretic Capacity Region

• Predicts the maximum wealth growth rate achievable by LLM agents under market coupling
• Frames multi-agent economic interaction as an information-theoretic channel
• Capacity region defines the boundary of achievable economic outcomes

Mean-Field Residual-Attractor Dynamics

• Models the dynamics of LLM agent economies using mean-field theory
• Residual attractor: the system converges to predictable equilibrium despite agent heterogeneity
• Connects microscopic agent behavior to macroscopic economic patterns

Pre-Registered Experiment

• Predictions registered before data collection (reduces p-hacking risk)
• Uses frontier LLM (Claude Opus 4.8) as economic agents
• Small-economy setting enables controlled experimentation

Methodology

1. Agent Setup: Frontier LLM agents act as economic actors in a market simulation
2. Market Coupling: Agents interact through market mechanisms (trade, pricing, etc.)
3. Wealth Tracking: Measure wealth distribution and growth across agents
4. Information-Theoretic Analysis: Test whether growth rates respect predicted capacity bounds
5. Dynamics Analysis: Fit mean-field residual-attractor model to observed dynamics
6. Pre-Registration: Hypotheses and analysis plan registered before experiment

Implications

• LLM agents as a testbed for economic theory: controlled, reproducible experiments
• Information-theoretic framework for understanding multi-agent economic systems
• Mean-field theory as a bridge between individual agent behavior and system-level dynamics
• Pre-registered methodology sets a standard for experimental rigor in agentic AI research
• Insights into how frontier LLMs behave in economic settings (rationality, strategy, cooperation)

Pitfalls

• Small economy size may not generalize to larger multi-agent settings
• LLM agent behavior may not reflect human economic behavior
• Pre-registration is valuable but doesn't eliminate all experimental biases
• Mean-field approximations may break down with heterogeneous agents
• Single LLM (Claude Opus 4.8) limits generalizability across model families
• Economic simulation design (market rules, initial conditions) strongly affects outcomes

Activation Keywords

LLM agent economies, information-theoretic capacity, mean-field dynamics, attractor dynamics, market coupling, multi-agent simulation, wealth growth, economic simulation, pre-registered experiment

Paper Reference

arXiv:2607.06001 - "Information Limits and Attractor Dynamics in Economies of Frontier LLM Agents" (Jul 2026)