BMT-04.01 Executive Summary#
BlueMirror.tech | May 2026#
Nadia spent eight years at health technology companies watching three generations of “patient empowerment” products fail identically. The first gave patients data. The second gave recommendations. The third gave AI-driven decisions without telling people how much the system had decided for them. When she joined BlueMirror, her first question was not how much autonomy the system has, but how the person knows how much autonomy the system has, and how she changes it.
The answer is the Human Agency Scale: a 0.0-to-1.0 spectrum that determines how much the system can do without asking. At 0.0, the system does nothing without instruction. At 1.0, it handles everything and reports back. Real life happens between 0.3 and 0.8, and the right setting varies by domain. Margaret wants medication reminders fully automated, appointment scheduling to require confirmation, financial decisions advisory only, and entertainment recommendations to do whatever they want. A single on/off switch cannot express this. A scale with domain modifiers can.
Six named levels have distinct behavioral signatures. Full Manual (0.0) means the system waits for explicit instructions. Informed (0.2) means it monitors and surfaces information without recommending. Advisory (0.4) means it recommends and waits. Guided Automation (0.6) means it acts within pre-approved patterns and notifies afterward. Trusted Automation (0.8) handles most decisions and surfaces exceptions only. Full Delegation (1.0) handles everything with periodic summaries and is never the effective level for healthcare or finance.
The critical innovation is the domain modifier system. Nine modifiers adjust effective autonomy from the person’s stated base. Healthcare carries a -0.3 modifier: a person at 0.7 overall has an effective 0.4 in healthcare. Legal carries -0.4. Financial carries -0.2. Entertainment carries +0.2. The modifiers are defaults, not mandates. A person who wants 0.8 in healthcare after two years of experience can set it. The override is her deliberate choice, visible and changeable.
The HAS is configured at onboarding through concrete behavioral descriptions rather than numerical settings. The first thirty days operate conservatively regardless of stated preferences. The system learns through behavioral signals which recommendations the person accepts, overrides, or ignores. After the initial period, the system proposes autonomy adjustments based on demonstrated competence, explicitly rather than silently.
The bidirectional principle means the system that earns the right to do more must also recognize when the person earns the right to do more herself. A person who starts managing her own appointments after the system handled them does not receive a warning or a confirmation prompt. The system asks whether she wants it to stop, and either answer is equally acceptable. A system that gradually takes over to make itself indispensable is creating fragility, not serving the person.
A third modifier operates in real time: the cognitive state modifier. When the Cognitive State Estimator detects reduced function, the effective HAS level adjusts downward automatically. The person experiences a system that seems to need a little more from her today. The adjustment is not labeled. When cognitive function recovers, the effective level returns.
The full article is available at bluemirror.tech.