Researchers have proposed a framework for developing artificial intelligence agents that can assist in molecular science. These 'molecular LLM agents' must be able to perceive and reason about chemical objects in various formats, including symbolic strings, graphs, and experimental data. The authors outline two complementary perspectives: an architectural design for the agents and a scientific autonomy ladder categorizing their capabilities. This framework aims to guide the development of future molecular LLM agents that can assist in discovery workflows. --- Why it matters: This work matters because it provides a comprehensive framework for designing and evaluating AI agents in molecular science, which could accelerate breakthroughs in fields like medicine and materials science. Source: https://arxiv.org/abs/2608.23104