Accountability
The obligation to answer for the outcome and carry the consequences of work undertaken under one’s authority.
Accountability does not disappear through delegation. Nor does taking proper care beforehand remove the obligation to acknowledge harm, repair what can be repaired, and answer when something goes wrong.
In practice
An agent makes an error despite reasonable precautions. The principal explains what happened and addresses the harm rather than treating the machine as the party that must answer.
Adoption
The process of bringing a capability into practical use and developing the habits, skills, and arrangements needed to use it well.
Access to a tool does not by itself constitute adoption. Adoption involves changes in practice.
In practice
A team learns to use an AI assistant, changes how work is reviewed, and checks whether it improves the work.
Agent
One who acts for another, within a mandate.
An agent may be human or artificial. Capability describes what it can do; its mandate determines what it may do. An AI agent does not become a principal by becoming more capable.
In practice
An assistant drafts a proposal under instructions. The ability to send it does not itself grant permission to send it.
Authority
The legitimate standing to direct, decide, or permit action within a defined scope.
Authority is distinct from technical capability. It can be exercised through delegation, but every mandate must trace to legitimate human direction.
In practice
A system can process a payment, but the authority to spend comes from an authorized human decision or mandate.
Autonomy
The discretion an agent has to act without seeking approval for each step, within a mandate.
Greater autonomy does not make an agent a principal or release the principal from accountability.
In practice
An agent handles routine scheduling independently and refers conflicts outside its instructions to a human.
Capability
The practical ability to perform an action or produce a result.
Being able to do something does not establish authority to do it. Capacity concerns the means available to exercise that ability under present conditions.
In practice
An AI can draft and send a message. Its mandate may permit drafting only.
Capacity
The means available to exercise an ability in a given situation, including time, attention, resources, and support.
Capability concerns what someone or something can do. Capacity concerns whether—and to what extent—they can actually do it under present conditions.
In practice
A person can assess an agent’s decisions but cannot review thousands each hour. The work must remain within their capacity for meaningful oversight.
Care
Earnest attention to what the work may affect, the risks it creates, and how harm can be reduced throughout its course.
Care operates within the bounds of respect. Some risks must be refused; others can be reduced and monitored. Care cannot guarantee that nothing will go wrong.
In practice
Before automating a process, examine who could be affected, keep failures contained, and reconsider the approach as new facts emerge.
Delegation
Entrusting work or decisions to an agent within a purpose and bounds set by a principal.
Delegation allows others to act. It does not transfer away the principal’s accountability or excuse losing the means to oversee the work.
In practice
Routine requests proceed within agreed limits; exceptions reach someone with authority to decide.
Dignity
The inherent worth of every human, independent of capability, usefulness, or status.
A person’s standing does not depend on outperforming a machine.
In practice
A person affected by an automated decision deserves respect and an answer even when they cannot understand or challenge the technology unaided.
Governance
The arrangements and practices through which people establish direction, assign authority, oversee its exercise, and hold those who exercise it accountable.
Governance establishes how decisions are made and answered for; management organizes work within that framework.
In practice
An organization defines who may authorize AI use, what limits apply, and who must answer for the outcomes.
In charge
Holding authority to direct work, exercising the responsibility to guide it, and remaining accountable for its outcome.
Being in charge does not require doing everything or approving every action. It requires meaningful human direction.
In practice
A principal delegates routine work while retaining the understanding and means to redirect it.
Judgment
The human exercise of understanding a situation, weighing what matters, and deciding how to act.
Rules, evidence, and AI analysis support judgment. They do not eliminate the need for someone to decide where rules are incomplete or competing considerations must be weighed.
In practice
A governance system flags a departure from procedure. The authorized people consider the circumstances and decide what follows.
Legitimacy
The justified basis for a claim to authority, grounded in the rights, mandates, and obligations that apply.
Having power—or being technically able to act—does not by itself establish a right to direct others.
In practice
Administrative access lets someone change a system, but their role and mandate determine which changes they are entitled to make.
Mandate
The purpose, scope, and limits within which an agent is authorized to act.
A desired outcome does not authorize every means of achieving it. A mandate must remain within the authority of the person granting it.
In practice
An agent may compare suppliers within a budget but must obtain approval before placing an order.
Oversight
The continuing practice of keeping delegated work visible, assessing it, and intervening when needed.
Oversight requires a practical ability to act; receiving reports alone is insufficient. It need not involve approving every step.
In practice
A principal reviews exceptions, checks outcomes, and can pause an agent when its work departs from its mandate.
Principal
The human for whom an agent acts, whose intent and authority give the work its mandate.
The principal sets the purpose and bounds. This is a role in a relationship: a human may act as an agent for someone else while directing agents of their own. Delegating the work does not remove the principal’s obligation to answer for its outcome.
In practice
A person asks an AI agent to research options within a stated scope. The person remains the principal.
Principle
A foundational commitment that guides judgment and action across situations.
A principle provides direction; a rule specifies what to do in defined circumstances. Principal names the person for whom an agent acts.
In practice
Human authority guides the design of a process even when no existing rule covers a new use of AI.
Purpose
The human end that gives work its direction and a basis for judging its value.
Completing a task or improving a metric does not necessarily serve its purpose.
In practice
A support team uses AI to help people resolve problems. Closing more tickets is useful only when it serves that purpose.
Respect
The active appreciation of what people value, including their consent, boundaries, and standing.
Respect sets limits on the purposes we pursue and the means we use. Toward ourselves, it also means acting coherently with our enduring values.
In practice
A useful result does not entitle an agent to enter another person’s systems or take their attention without consent.
Responsibility
The capacity and ongoing duty to perceive what is happening, exercise judgment, and act while acting still matters.
Responsibility concerns our ability and ongoing duty to respond. Accountability concerns answering for outcomes. A title alone does not create the practical conditions needed to exercise responsibility.
In practice
A principal sets limits, keeps the work visible, and can intervene before an error becomes irreversible. They need not approve every routine action.
