BMT-01.16 Executive Summary#
BlueMirror.tech | May 2026#
Margaret Chen has not told her daughter what she decided last Tuesday. She has not told her cardiologist either, though the cardiologist’s visit three weeks ago was part of what brought her to it. She sat with it for several days before she was ready to name it, even to herself: she is not going to pursue the aggressive treatment option. She is seventy-three. She has had a good life. The treatment would extend the quantity of the time remaining. She is more interested in the quality of it. The decision is hers. She is thinking about how to name it to the people who love her, in what order, in what way.
The end-of-life concierge does not activate at a diagnosis. It does not activate when a clinical signal crosses a threshold. It activates at a decision: the person, on her own terms, indicates that her frame has changed. That the question she is now asking is not how to maintain function or extend independence but how to define and protect the life she wants in the time she has. From that decision forward, the architecture follows her. Not a clinical protocol. Not a family consensus. Her.
Every other concierge agent in the system is organized around capability. The health concierge monitors function and coordinates care because function matters. The cognitive concierge scaffolds independence because independence is worth scaffolding. The end-of-life concierge marks the moment when the person changes what matters most to her: from capability to comfort, from quantity to quality, from future-oriented to present-oriented. This is not a collapse. Many people who have made end-of-life decisions are living full, active, engaged lives. What the palliative frame changes is the hierarchy of trade-offs.
Four infrastructure agents power the end-of-life concierge, and all four are dormant until activation. The Advance Directive Manager holds advance directives and healthcare proxy designations in structured, versioned storage, monitors for revision requests, and coordinates with the legal advocate for documentation. Its autonomy default is 0.25, the lowest in the system; no action proceeds on any directive without explicit subscriber or authorized proxy confirmation. The Comfort Monitoring Agent observes and records symptom and pain patterns through self-report and behavioral inference. It does not produce clinical assessments; it documents and surfaces patterns to the subscriber and, with her authorization, to designated clinical contacts. The Legacy Coordination Agent manages the capture and controlled transmission of legacy assets: letters, recordings, personal histories, instructions she wants carried out. No transmission occurs without explicit subscriber instruction or a post-death trigger the subscriber has pre-configured and a designated proxy has confirmed. The Care Circle Transition Agent manages the shift in care circle communication from routine coordination patterns to end-of-life context patterns, working within permissions already granted and requiring subscriber confirmation for each change.
Activation requires explicit subscriber decision, or authorized proxy action within the scope the subscriber defined in advance. The system can observe clinical signals that suggest end-of-life considerations may be relevant. It can surface the option. It cannot enter the frame without the person’s decision. Deactivation is equally available: the person who entered the frame and then decided to exit it can do so. The system follows the person in both directions. The activation model is not a one-time grant. It is continuous sovereignty.
Three refusal conditions are absolute. The concierge does not activate without explicit authorization. The system cannot be activated by a clinical signal, a family member’s wish, or a system inference. Only the subscriber or a designated proxy operating within subscriber-defined scope. It does not communicate end-of-life status to family members without per-recipient subscriber authorization. The daughter’s wish to know does not entitle the daughter to know; the mother’s authorization does. It does not substitute clinical palliative care judgment. When clinical interpretation is required, the concierge connects the subscriber to the clinical professionals who provide it. It coordinates and supports. It does not assess prognosis or recommend care pathways.
The ethical weight this agent carries is proportionate to what is at stake. The consent architecture required for this agent extends into territory the standard framework does not cover: the conditions under which advance directives govern when the subscriber can no longer direct her own care, the scope and limits of proxy authorization, the escalation hierarchy for palliative-specific scenarios. The internal build specification for this agent is the most complex per-agent document in the corpus. The weight is the architecture taking this seriously. The architecture holds the complexity and serves her within it, because a person approaching the end of her life deserves a system that does not simplify what is genuinely complex.
What is left: Margaret’s Wednesday in April, which opened this series, runs from 6:14 a.m. to 10:18 p.m. Fifteen agents work on her behalf. Thirteen work toward capability, connection, and the life she is building. One learns her and helps her understand the world she is navigating. One waits, ready, for a different conversation. She has not needed that conversation yet. When she does, the architecture will be there.
For the full treatment of the frame that changes, the four infrastructure agents, the activation model, and the ethical architecture, read the complete article on BlueMirror.tech.