Researchers introduced the Qiushi Discovery Engine, an LLM-based system that conducts autonomous scientific research on real optical hardware without human guidance beyond an initial goal. Over a run involving 145.9 million tokens, 3,242 LLM calls, and 1,242 tool calls, Qiushi reproduced a published transmission-matrix experiment on a new platform, produced the first experimental observation of a coherence-order structure class, and discovered a previously unreported optical bilinear interaction the authors describe as "structurally analogous to a core operation in Transformer attention"—with potential applications in energy-efficient optical hardware for pairwise computation. The authors claim this is the first instance of an AI agent autonomously identifying and experimentally validating a nontrivial, previously unknown physical mechanism.