Multi-agent framework lets LLM agents design and run controlled experiments with simulation models

A team of researchers has proposed a multi-agent framework that enables large language model (LLM) agents to conduct controlled experiments using scientific simulation models, specifically for pharmaceutical process design. The system takes a user query and a baseline configuration, then constructs a structured task representation, designs experiments, executes comparative simulations, interprets outcomes, and synthesizes evidence-based recommendations for process parameter optimization.

The framework integrates LLMs with high-fidelity simulation models within an interactive agent loop, allowing the agents to reason through intervention, comparison, and observation rather than relying solely on text or code generation. The authors report that this approach produces more specific and actionable outputs than language-only reasoning.

In an industrial application setting, the system’s outputs were rated higher by users in terms of specificity, correctness, and helpfulness compared to language-only reasoning. The paper also includes ablation studies and visualized case analyses to demonstrate the system’s effectiveness and practical utility.

The work was submitted to the 31st IEEE International Conference on Emerging Technologies and Factory Automation (ETFA 2026) and is available as an arXiv preprint (arXiv:2608.23622). The research team includes authors from academic and industrial backgrounds, with affiliations including the University of Stuttgart and AstraZeneca.