Researchers at Stanford have built an agentic artificial intelligence platform called MetaChat that radically accelerates the design of advanced photonic devices, such as metasurfaces, that is, extremely thin structures that control light at the nanoscale, tells Stanford Report. The traditional design of these optical components requires expert knowledge, heavy simulations, and weeks or months of trial and error. MetaChat combines self-reflective AI agents with high-speed simulation tools to cut that timeline to minutes while collaborating with human designers.
Metasurfaces have applications in imaging, sensing, augmented reality, and optical computing because they can bend, focus, or shape light in ways that traditional lenses cannot. But creating these structures means solving Maxwell’s equations, which govern electromagnetic fields, thousands of times during iterative design cycles. The Stanford team solved this bottleneck by developing FiLM WaveY-Net, a deep-learning neural network that can simulate electromagnetic behavior over a thousand times faster than conventional methods, bringing simulation times down to milliseconds.
MetaChat’s innovation lies in its use of AI agents with agency, meaning they can make design decisions, reflect on outcomes, correct their strategies, and query human users when needed. One agent acts like an optics designer and another like a materials expert, pulling data from databases and running fast simulations. Instead of rigid workflows, the agents engage in iterative reasoning, enabling real-time exploration of design options. In tests, MetaChat produced a complete, downloadable design for a multi-wavelength metalens in just 11 minutes, a task that normally takes human engineers much longer.
The tool provides a chat interface where engineers can describe goals in natural language, and the AI generates optimized photonic layouts accordingly. Researchers say this democratizes access to specialized optical design knowledge and could help address the shortage of skilled optical engineers. While MetaChat won’t replace human insight, it dramatically expands what teams can explore and create, opening doors to faster innovation across sensing, imaging, and photonics technologies.