
Explore Discover, the Knowledge Hub, playbooks, and the Operator Vault.
Glossary
Human-in-the-loop (HITL) means a human participates in or must approve a step inside an operating decision workflow before the system proceeds — for example confirming a refund, signing off a configuration change, or choosing among agent-proposed actions. It is distinct from human-on-the-loop supervisory monitoring of automated operation, and from retrospective human auditing after actions have already completed.
Agentic systems create speed; HITL creates intentional friction where decision rights demand it. Without clear placement of human checkpoints, organizations either slow everything with blanket approvals or discover after incidents that no one was accountable at the moment of action.
Related oversight patterns should not be collapsed into one label: - Human-in-the-loop — approval or participation integrated into the live workflow (the system waits or branches on a human decision). - Human-on-the-loop — humans supervise dashboards, alerts, or kill switches while automation continues unless interrupted. - Retrospective auditing — sampling, log review, or compliance checks after the fact; valuable for governance, but not the same as mandatory in-loop approval.
In the TSS Agentic Operating Model, HITL maps to decision rights: which actions an agent may complete alone, which require dual control, and which remain human-owned.
Example (enterprise): A procurement agent can draft a vendor request and gather quotes autonomously, but purchase orders above a threshold enter HITL approval by finance. Separately, operations may sample completed low-value orders for audit quality — that sampling supports oversight without being an in-loop gate for every order.
None linked yet.
No related resources yet.