---
type: "synthesis"
clusters: ["C2", "C4", "C7"]
id: "meta-human-in-the-loop-standard"
sources: ["cross-day"]
---
Across Agentic, Governance, Adoption, and the Tail, one architecture recurs: automate the routine, escalate the exceptions to a named human. Agentic: [[concept-human-in-the-loop-escalation]], engineered [[action-design-hesitation|hesitation]], the [[concept-orchestration-layer]] ([[cross-hitl-escalation]]). Governance/Trust: the [[framework-trustworthy-ai-triad]] and [[concept-ai-fiduciary-duty]]. Tail: 'algorithms suggest, humans decide' ([[cross-algorithm-as-guide-human-judgment]], [[action-empower-frontline-managers]]). The genuine tension is the oversight paradox: push humans to verification and verification becomes the new bottleneck ([[cross-oversight-paradox]], [[concept-oversight-capacity]], [[question-verification-bottleneck]]); over-supervising a consumer agent also defeats its purpose ([[contrarian-supervision-defeats-ai]]). Resolution the corpus reaches: **risk-differentiated** oversight — automated safeguards absorb most validation, humans hold the liable exceptions. Companion: [[meta-decision-architecture-reboot]], [[meta-judgment-the-new-scarcity]].