There is no standard async pause and resume mechanism for AI agent human approval flows

Most teams add human-in-the-loop as an afterthought with no architecture that can pause agent execution, wait for a human, and resume from the same state without corrupting context. Naive implementations create decision fatigue bottlenecks where reviewers are flooded with low-risk approvals. HumanLayer, a YC startup building exactly this, launched with 47 HN upvotes. Ottr launched a one-click approval link approach. The EU AI Act making human oversight mandatory for high-risk agents from August 2026 is forcing this issue into every enterprise agentic deployment. No framework has a standard approval protocol.