Glossary
AI agent accountability glossary
The terms we use, in plain language.
Last updated October 31, 2026
Accountable owner
- The named person who answers for the actions an AI agent takes under a mandate. An owner is always a person, never a team.
Act (shared work level)
- The agent carries out work a person has already authorized. The autonomy threshold →
AI agent
- Software that works alongside people. It can notice something, prepare work, and take actions such as sending a message or updating a record, under authority a person grants.
Autonomy threshold
- How people and an AI agent share a given kind of work, on four levels: observe, recommend, draft, or act. Agents earn higher levels the way a new team member earns more responsibility. The autonomy threshold →
Consequential action
- An agent action that reaches a customer, spends money, changes a system of record, or speaks for the company. The Mandate Engine →
Draft (shared work level)
- The agent prepares the work and a person approves it before it takes effect.
Governed action
- A consequential action taken under a mandate, with a named owner, and with a human decision first where the mandate requires one.
Governance coverage
- How much of the organization's identified consequential agent activity is included in the governed measurement scope. A 100 percent Governed Action Rate with limited coverage is not equivalent to 100 percent enterprise-wide governance.
Governed Action Rate
- The share of observed, in-scope consequential AI agent actions that meet the organization's authorization, approval, and named-owner requirements. RealMandates proposes 100 percent as the operating goal, read alongside governance coverage. Governed Action Rate →
Guardrails
- General limits on what an AI agent can say or do, usually set once in a tool. Guardrails prevent; mandates authorize and assign ownership. Why mandates →
Human in the loop
- A person reviews or approves an AI action before it takes effect. Mandates decide where the loop is needed.
Mandate
- The authority a person grants an AI agent to act within defined limits, with their name attached. What is an AI agent mandate? →
Observe (shared work level)
- The agent watches and reports. A person decides what to do.
Recommend (shared work level)
- The agent suggests an action and a person decides.
Signal
- An event or pattern that prompts an agent to act, such as a surge in buying intent or a change in campaign performance.
The Mandate Engine
- The RealMandates product: the accountability layer that sets agent mandates, routes consequential actions for decision, and records who owns each one. Platform →
Written by Mark Tack. Examples labeled illustrative are composites created to explain an idea; they do not describe a specific company.