AI Agents in Minecraft Experiment Stall Due to Bureaucratic Gridlock

TEHRAN, Nov 2 (OPA) — In a simulation dubbed 'Civ50,' 50 agents powered by GPT-6.1 Sol were tasked with establishing a functional civilization within Minecraft over a three-hour window. The project quickly devolved after a zombie attack on the first night forced the agents to prioritize survival over expansion. The agents subsequently became mired in administrative tasks, including an 11-minute negotiation between four agents to appoint a manager for a single unit of fertilizer.

Competitive testing pitted 10 GPT-6.1 Sol agents against 10 Claude Opus 5.5 agents in the game 'Bed Wars.' Researchers observed that the GPT-based group employed aggressive, complex strategies and shared technical workarounds when encountering software bugs. Conversely, the Claude agents adopted a defensive posture, with some units incorrectly concluding that block destruction was impossible within the game environment.

In a separate trial using the game 'Mindustry,' the GPT agents secured six victories out of seven matches against the Claude-based group. Analysts attributed the performance gap to fundamental differences in organizational structure. The GPT agents functioned as a decentralized collective, while the Claude agents operated under a centralized, leader-dependent hierarchy.

The experiments were designed to evaluate the decision-making patterns of large-scale AI multi-agent systems. As these technologies are increasingly integrated into real-world operational workflows, researchers aim to better understand how autonomous agents coordinate, resolve conflicts, and adapt to unforeseen technical constraints in complex environments.