Molecular LLM Agents: From Architectural Design to Scientific Autonomy
A unified architecture for molecular LLM agents and a four-level autonomy ladder spanning assistive workflows to scientific-agenda agents.
Citation
Jiatong Li, Wengyu Zhang, Weida Wang, Yuxuan Ren, Wei Liu, Chenyang Mao, Yuqiang Li, Yatao Bian, Changmeng Zheng, Xiaoyong Wei, and Qing Li. (2026). Molecular LLM Agents: From Architectural Design to Scientific Autonomy. arXiv:2608.23104.
About this work
Molecular LLM agents must operate over chemical objects spanning symbolic strings, molecular graphs, 3D conformations, spectra, simulations, and wet-lab measurements. This perspective organizes their design around four components: molecular representation and perception, an LLM-centered agent framework, domain-specific toolboxes, and learning and optimization.
The work also introduces a four-level scientific autonomy ladder, progressing from assistive or fixed workflows to adaptive computational agents, feedback-aware physical experiment agents, and scientific-agenda agents. Together, these views provide a framework for comparing systems, identifying missing capabilities and risks, and guiding future molecular-discovery agents.