InquirySpec - Ontological Boundary: Supported Agency means agents execute situated judgment inside scaffolds that carry routing and governance burdens. - Not This: Not autonomy hype, learned helplessness, or central command. - Doctrine Dependencies: Workflow Engine, lC.Core, Accountability_Assessment.
Working Definition
Supported Agency is the design principle that lets bounded actors act without pretending to be complete systems. A person, AI agent, team, or institution can make situated judgments, but it cannot simultaneously own the whole mission frame, remember every prior state, infer the workflow, enforce policy, validate evidence, route tools, preserve records, and close the work with accountable authority. When all of that is loaded onto one actor, agency does not become stronger. It becomes distorted by overload.
Supported Agency therefore separates the proper burden of the actor from the burdens that should be carried by the field around the actor. The actor still observes, interprets, chooses, adapts, and answers for their local work. The support field carries state, routing, memory, governance, verification, escalation, and controlled release. The result is neither command-and-control nor learned helplessness. It is scaffolded agency: enough structure to prevent collapse, enough room for judgment to remain real.
The Phenomenological Problem
Unsupported work usually fails quietly before it fails dramatically. A team enters a meeting with a dashboard, a transcript, a ticket queue, a policy memo, and a deadline. Each artifact contains a partial signal. None of them carries the full ecology that produced it. The person in the room is then expected to integrate the signals, remember the history, infer the authority boundary, judge the evidence, decide the next move, and phrase the decision in a way the institution can accept.
That is the monolithic burden in ordinary clothes. It is not a villain plot. It is the metabolic tax of complex coordination. Under pressure, systems drift toward the path of least resistance: the visible metric, the nearest precedent, the most fluent explanation, the action that keeps the workflow moving. Even well-meaning people can become unscaffoldedly disingenuous when the environment gives them no reliable way to carry context, disagreement, uncertainty, and consequence at the same time.
Supported Agency names the alternative. The point is not to remove responsibility from the actor. The point is to stop asking the actor to carry burdens that should have been designed into the environment.
The Engineering Anchor
Inside the learnt.cloud architecture, Supported Agency is anchored by the Workflow Engine. The engine rejects the expectation that a single autonomous agent should define scope, plan the route, remember total state, apply governance, execute the work, validate the result, mutate canonical memory, and synthesize final meaning. Those burdens are separated across the surrounding cyber-physical field.
The internal service layer gives that separation a mechanical shape. Input and output boundaries control ingress and release. Process routing determines what kind of step is being taken. State services orient the actor inside the current workflow. Governance applies policy and closure rules. Memory preserves context and evidence. Verification checks whether the artifact survives contact with declared constraints. In public terms, the system helps the actor know where they are, what they are allowed to touch, what evidence matters, what must be preserved, and when the work has reached a legitimate stopping point.
AAC adds the accountability boundary. A support field may record evidence and diagnose whether accountability structures are functioning, but it does not become the forum of judgment. The machine is not absolution. The forum is not automatically just. The actor is not erased by systemic pressure. Supported Agency holds these together: it relieves impossible coordination load while preserving answerability for situated choices.
Boundary Conditions
Supported Agency is not autonomy hype. If a system says the agent can do everything, it has usually hidden the burdens rather than solved them. That failure belongs with the Monolithic Agent Fallacy.
Supported Agency is not central command. The scaffold should not make every local judgment wait for a distant authority. It should make the boundary conditions explicit enough that local judgment can happen safely.
Supported Agency is not moral outsourcing. A support field can reduce confusion, preserve records, and expose consequence patterns, but it cannot choose who a person becomes inside a coercive maze. Accountability remains distributed across evidence, forum, system design, and agent-level action.
Supported Agency is not total transparency. The field must preserve assessable traces at the right fidelity, but routine work does not require every internal trace to be exposed to every audience. Reviewability is governed by purpose, risk, policy, and trust maturity.
Drill Path
- Workflow Engine explains the coordination layer that distributes state, governance, routing, verification, memory, and release.
- Monolithic Agent Fallacy names the failure mode created when one actor is expected to carry the whole system.
- Air-Gapped Support Field explains why separation, staging, and governed transfer protect judgment from premature synthesis.