Agentic Care Coordination for Chronic Conditions
Designing ethical constraints and failure architecture for autonomous health management
The Challenge
A healthcare provider designed an AI agent system to coordinate ongoing care for patients with chronic conditions - managing medication schedules, monitoring symptom patterns, coordinating between specialists, and adjusting care plans within clinician-defined parameters. Healthcare presents the most demanding trust architecture challenge in AXD: the consequences of agent failure are not financial but physical. A medication timing error is not a refund scenario - it is a patient safety event. The design had to embed clinical governance into every layer of the agent's authority.
AXD Approach
- ■Designed a clinician-delegated authority model: the AI agent operated exclusively within parameters set by the treating clinician, with no capacity for self-expansion of authority - every boundary was a clinical decision, not a technical default
- ■Implemented a three-tier safety architecture: Tier 1 (autonomous) covered routine scheduling and reminders, Tier 2 (supervised) handled dosage adjustments within pre-approved ranges, Tier 3 (clinician-only) covered any action outside established parameters - with automatic escalation and no override capability
- ■Built an Ethical Constraints framework specific to healthcare: the agent could not withhold information from patients, could not prioritise efficiency over patient preference, and was required to explain its reasoning in patient-accessible language for every recommendation
- ■Created a failure architecture with zero-tolerance for silent failures: any missed medication reminder, any unacknowledged symptom report, any failed specialist coordination triggered immediate escalation to a human care coordinator with full context
- ■Designed patient consent as continuous and granular: patients could modify their delegation at any time, withdraw consent for specific agent functions while maintaining others, and access a complete audit trail of every action the agent took on their behalf
AXD Principles Applied
- ◆Founding Principle 1: Agency Requires Intentional Delegation - clinical delegation was explicit, bounded, and revocable, with the clinician as the delegating authority
- ◆Founding Principle 2: Trust is the Primary Material - trust architecture in healthcare required zero tolerance for undetected failures, making safety the primary trust signal
- ◆Ethical Constraints Framework - healthcare demanded the most rigorous application of ethical boundaries in any AXD implementation
Design Outcomes
- →Clinician-delegated authority prevented the agent from exceeding its clinical mandate, maintaining professional governance over autonomous actions
- →Three-tier safety architecture ensured that the severity of potential consequences determined the level of human oversight required
- →Zero-tolerance failure architecture eliminated silent failures, ensuring every anomaly was surfaced to human care coordinators
- →Continuous granular consent gave patients meaningful control over their care coordination without requiring them to understand the technical system
Key AXD Insight
Healthcare is where AXD's founding principle - that trust is the primary material - faces its most demanding test. An agent that manages medication for a patient who is asleep is the purest expression of absent-state design. The design must be so robust that the patient's trust is warranted even when they cannot verify it. In healthcare, trust architecture is not a design preference - it is a clinical obligation.
Related Reading
Frequently Asked Questions
How does AXD apply to healthcare?
AXD applies to healthcare through clinician-delegated authority models, three-tier safety architectures, ethical constraints frameworks, and zero-tolerance failure design - ensuring autonomous care coordination maintains clinical governance and patient safety.
What is zero-tolerance failure architecture?
Zero-tolerance failure architecture means any missed action, unacknowledged report, or failed coordination triggers immediate escalation to a human care coordinator with full context - eliminating the possibility of silent failures in safety-critical autonomous systems.
Apply These Principles
This case study illustrates AXD principles in context. To apply them to your own organisation, start with the AXD Readiness Assessment, explore the 12 frameworks in The Practice, or consult the AXD Playbook for a structured implementation guide.