Autonomous Artificial Worlds with Self-Sustaining Economies: Autonomous Agents, Artificial Markets, and Social Simulations — A Comparative Survey of 77 Systems, Research Foundations, and Research Lineages

<Click here for a list of content prior to July 22, 2026>

1. Why a map of 'worlds with self-sustaining economies' is needed

Experiments involving dozens or hundreds of large language model agents have rapidly become commonplace over the past two years. Giving agents roles, providing them with memory, and having them negotiate with each other—all of this can now be done on a personal computer.

However, when you try to take the next step and build a system where 'a large number of entities reside in the same world, exchange resources, prices fluctuate, some make a profit, and others go bankrupt,' the precedents to refer to suddenly become invisible. While there is a vast amount of literature on multi-agent systems, most of it consists of experiments with dozens of agents and hundreds of steps, rather than discussions on sustaining the world itself.

In reality, however, there is thirty years of accumulated knowledge in this type of system. Agent-based modeling (ABM), artificial markets, macroeconomic simulations, urban and traffic simulations, and—most overlooked of all—commercial games.

This article is an attempt to map that accumulation. We have collected 77 cases corresponding to 'artificial worlds with self-sustaining economies,' primarily from Europe, the United States, and East Asia, scored them along eight common axes, and organized them by region, type, and openness. The intended readers are engineers designing multi-agent systems, researchers in computational social science, and social science readers interested in institutional design and policy simulation.

I will state one of the conclusions first. Among the top 10 systems with the highest degree of economic self-sustainability based on these eight axes, 9 were games or implementations close to games, 8 were commercial licenses, and not a single one was fully open-source software (OSS). However, as will be discussed later, this result depends on how the indicators were constructed and cannot be interpreted simply as 'games are superior to academia.'