A 38-Page Survey Proposes "ComBodied Agents" — Making the Human's Evolving State, Not Software or Physical State, the Agent's Primary Object

arXiv / HuggingFace Daily Papers·medium signalXBlueskyLinkedInCopy link

Twenty-two authors including Dacheng Tao and Xingyao Wang posted a 38-page position paper on August 11 (2608.10915, top of HuggingFace Daily Papers at 118 upvotes) arguing current agent research has a structural gap: "Digital Agents primarily transform software states, while Embodied Agents transform physical states; neither makes a person's evolving state and agency the primary object." Their proposed paradigm combines multimodal perception, longitudinal memory, predictive modeling of the individual, and intervention policies to track and support a specific human's trajectory over time. It's a taxonomy rather than a system — 6 figures, 10 tables, no benchmark — but it's the clearest articulation yet of why personal-agent products keep failing to compound.