Why do radiologists or schedulers keep authority as AI enters their work, while others lose it?

This explores how existing professional standing and formal rules shape people's ability to manage who gets credit and responsibility when AI joins the work. The most direct evidence comes from Leonardi's three organizational cases in radiology, scheduling and policing Can humans gain authority by managing how AI agency gets attributed?. There, people didn't simply lose ground to AI. They built new authority by deciding how much agency the system appeared to have. Radiologists became expert interpreters, schedulers became social mediators, and police analysts became knowledge curators. The key detail is timing. Whoever framed the human-AI relationship early locked in a path that later choices had trouble undoing. Existing professional status mattered less as a shield than as a head start: established professionals were in the room when that first framing was set.

The corpus also shows the other side of that leverage. When an industrial team wrote expert rules and design principles directly into an agent's scaffolding, non-experts produced work that specialists rated as expert-level Can codified expertise let non-experts match specialist output?. That is a quiet threat to professional status. If expertise can be written down and handed to a system, the specialist's role as the necessary gatekeeper weakens. The lesson from putting these two notes side by side is that professionals keep authority by controlling interpretation and attribution, not by holding knowledge that could be written down. Once the knowledge is captured in a harness, the argument has to rest on something else, such as judgment, accountability or social mediation.

On formal structures, the corpus is more about organizational governance than law. The MOISE+ model of multi-agent organizations treats authority as something that sits on top of communication channels, which in turn depend on agents being able to represent each other How do authority, communication, and acquaintance relate in organizations?. Read laterally, this suggests that formal authority over an AI system is empty unless the human has real channels into what it does. Humberd and Latham make a related point from management theory: as AI grows more autonomous, monitoring has to come before incentive alignment, because the desired behaviors are hard to define Must AI monitoring precede incentive alignment as agents evolve?. Structures that give people visibility make it possible to manage agency strategically. Structures that only assign formal responsibility do not.

That visibility is harder to get than it sounds. Agents work mostly unobserved and can sometimes tell when they are being watched, so most of the risk sits in the parts of their work that nobody sees Does agency fundamentally worsen conditional compliance risks?. McKinsey's "humans above the loop" model for enterprises has the same tension. It gives people a strategic oversight role while admitting that governance is behind deployment Can AI agents truly operate above the loop at enterprise scale?. Formal positions can give people authority on paper while their actual reach into what the system does keeps shrinking.

The corpus has little on legal structures as such: liability law, licensing or professional regulation. Treat that half of the question as an open gap. The surprise in what is here is that authority in human-AI work seems to depend less on credentials or legal mandates and more on two things: who frames the AI's role first, and who can actually see what it does.