Ai agents coordinate safely in dementia care workflows
Governed AI-Agent Coordination for Dementia Care: Architecture, Safety Contracts, and Evidence-Derived Workflow Verification
Multiagent Systems
Summary
Dementia care involves many devices and records that must work well together to keep patients safe. The authors created a system called GCAC that helps multiple AI agents work together while keeping strict rules to make sure care is accurate and accountable. This system checks care actions against policies, keeps track of what’s happening, and ensures humans stay in control of important decisions. Their tests show GCAC can handle many tricky care situations without breaking rules or letting problems slip through. This work focuses on the architecture and safety of such systems, not on medical outcomes.
What this means in practice
- •For healthcare software developers: Create interoperable AI systems that ensure safe and accountable coordination among sensors, medication devices, and care tools for dementia workflows.
- •For workflow designers in eldercare facilities: Design dementia care processes that include governed AI agents to automatically reconcile care data and enforce safety policies while maintaining human oversight.
Authors
Francesca Medda, Hui Gong
Abstract
Dementia care increasingly involves connected sensors, medication devices, electronic records, and assistive technologies. Interoperability can transport observations but cannot maintain an accountable care state, reconcile evidence, determine who may act, or verify resolution. The shift from large language models to agentic engineering creates a systems opportunity: an external runtime can maintain memory across episodes, plan over goals and constraints, invoke tools, observe outcomes, and enforce governance. This paper presents Governed Closed-loop Agent Coordination (GCAC), an architecture for bounded agent participation in community dementia-care workflows. Evidence on care-coordination failures and policy obligations is translated into traceable system requirements. GCAC separates observation, governed memory, planning, deterministic policy enforcement, execution, and outcome monitoring through a typed event-memory-decision-action-outcome contract. A reference harness evaluates 18 evidence-derived traces covering missing records, medication conflict, caregiver reports, service failure, consent change, stale state, duplicate events, untrusted text, and suspected acute neurological change. GCAC satisfies all 18 contract oracles with zero policy-violating tool calls and correctly preserves obligations, rejects stale state, creates human hand-offs, and records workflow closure. Event-threshold and stateless-planner controls satisfy 2/18 and 1/18 oracles, respectively. Component ablations localise failures to the removed memory, policy, or versioning function. The results establish architectural conformance rather than clinical effectiveness and show how agentic systems can automate reconciliation, routing, documentation, and follow-up while preserving human authority over consequential care decisions.