In a new research paper, Hugging Face researchers propose a paradigm shift for Agentic AI, introducing 'Combodied Agents' to address a structural gap in current systems. The paper argues that while digital agents transform software states and embodied agents transform physical states, neither adequately models the evolving state and agency of the individual they are meant to serve.
The motivating example involves an older adult missing a medication dose. A digital agent might send a reminder, and a robot could deliver the medication, but neither system can determine whether the person forgot, is confused, experienced side effects, or deliberately refused—nor what support would be appropriate. This highlights the need for a human-centered approach that makes the person's state the primary focus.
The proposed Combodied Agents framework is a closed loop that integrates event-based multimodal perception, longitudinal correctable memory, Personal World Models, and an admissible intervention policy. These components work together to estimate future personal states under different decisions, select proportionate support under constraints like consent and safety, and update based on feedback from the person and environment.
Rather than requiring an exhaustive digital twin, the framework uses purpose-bounded, uncertainty-aware, user-correctable representations. The paper organizes the design space by human-state targets, relational contexts, and agent roles, and suggests evaluation methods, agency-preservation metrics, benchmark requirements, edge-native personal models, and governance directions. The authors position Combodied Agents as shifting Agentic AI from external task completion toward sustained human benefit.