In Situ Training of Optical Neural Networks by Coding Agents Coding agents built on large language models can train optical neural networks in situ, according to research on a method that replaces expert-built calibrated digital device models and gradient-free trial and error. The approach addresses the fact that every optical neural network device departs from its design and must be trained with the hardware itself. The work matters because optical neural networks promise fast, energy-efficient computing but have been held back by the per-device calibration burden. Optical neural networks promise fast, energy-efficient computing, but every device departs from its design and must be trained with the hardware. In situ training relies on an expert-built calibrated digital model of each device or on gradient-free trial and error. Coding agents, large language mode