Humans should supervise agent loops from a leveraged point, not sit inside every one

Zhao challenges the reflexive “human-in-the-loop” virtue: programming agents “taught us that having a ‘human-in-the-loop’ isn’t always desirable. It’s like having someone personally inspect every bolt on a factory line, or walk in front of a car to clear the road” — a reference to Britain’s 1865 Red Flag Act, which required a flag bearer to walk ahead of every motor vehicle. “We want humans to supervise the loops from a leveraged point, not be in them.”

This sharpens the management shift in Parallel agents create a management problem, not a coding problem: the win isn’t reviewing each agent’s every step, it’s positioning yourself to oversee many loops at once — the “manager of infinite minds” stance behind Treat AI like a distributed team, not a single assistant, operationalized by an An orchestrator agent that manages other agents solves the parallel coordination problem without human bottleneck. The unsolved precondition is verifiability — Zhao concedes humans must stay in the loop for now precisely because general knowledge work can’t be auto-verified the way code can, which is why Production agents route routine cases through decision trees, reserving humans for complexity reserves human attention for genuine ambiguity rather than spending it on every routine pass.