The Policy Test Lab: An Agentic AI-Based Simulation Tool

Abstract

As an emerging technology, Agentic Artificial Intelligence or Large Language Models (LLM)–based/Generative AI agent systems, have been increasingly adopted to enable autonomous reasoning, tool-using, and decision-making systems that transcend the traditional boundaries and use cases of language models. Many of these systems rely on multiple interacting agents, each capable of perception, reasoning, and adaptive behavior. Such agents collaborate in multilayered ecosystems to achieve cognitive and operational objectives. The policy test system introduced in this paper extends the agentic paradigm into the policy simulation domain by designing an LLM-based orchestration agent that autonomously interprets academic documents and translates them into executable implementations of simulation and optimization models. While this research is still in its infancy, the agentic tool already extracts an analogous model from peer reviewed literature, where the LLM serves as a cognitive controller, parsing unstructured knowledge into executable code. The resulting code provides simulation agents with the underlying dynamics of policies, resource flows, and behavioral adaptation. This work is a step towards a future where generative reasoning agents autonomously analyze, simulate, and optimize complex socio-technical systems in support of informed policy exploration.