Google presents the idea of a Personal Health Agent (PHA) powered by large language models (LLMs). The PHA combines data from wearables, biomarkers, and user inputs to offer personalized, evidence-based health advice. It breaks its function into three specialist sub-agents: a data science agent that analyzes and interprets the user’s raw data, a domain expert agent that grounds recommendations in medical knowledge, and a health coach agent that guides behavior change through conversation. An orchestrator oversees which sub-agent leads at each step, allowing them to collaborate and refine findings. In evaluations on real user data, the PHA outperformed single “all-purpose” agents and naïve combinations of sub-agents. The architecture suggests a path toward future AI systems that better support personal health needs in a modular, collaborative way.
