Simulating Two-Sided Job Marketplaces with AI Agents

Silvia Terragni, Behnaz Nojavanasghari, Frank Yang, Andrew Rabinovich

Keywords: Large Language Models, AI Agents, Two-Sided Marketplaces, labor markets, Simulation Framework, agent-based computational economics

TL;DR: We simulate two-sided marketplaces with LLM agents to study how reasoning and strategic adaptation drive emergent economic behavior

Abstract:

We introduce a simulation framework for studying how artificial intelligence agents behave in economic marketplaces. Unlike traditional computer simulations that use predetermined rules, our approach uses large language models (LLMs) as intelligent agents that can make strategic decisions and adapt their behavior over time. The framework includes reputation systems that track agent performance and detailed logging of decision-making processes.

Through systematic experiments comparing different types of agents, we reveal three key insights: the reasoning capabilities of AI agents create fundamentally different market behaviors, successful marketplaces require compatible decision-making approaches from all participants, and no single performance measure captures market success—instead, there are important trade-offs between transaction volume and match quality. The framework captures both what happens in the market overall and why individual agents make specific decisions.

This work provides a reproducible research platform for investigating how AI decision-making affects economic outcomes in controlled virtual marketplaces. To support reproducibility and further research, we make the complete simulation framework and analysis tools publicly available as open-source software at \url{https://github.com/upwork/simploy}.

\textbf{Disclaimer:} This simulation is illustrative, not prescriptive, and highlights the capabilities and limitations of LLMs in synthetic environments. The framework is not designed to model or evaluate real-world labor platforms or populations, and findings should not be used to draw conclusions about existing economic systems.

Submission Number: 242