Kadambi et al. propose a dual-embodiment framework for multimodal large language models based on internal and external embodiment. Such biologically inspired principles may help inform next-generation AI systems.