From Hierarchy to Team: Designing Shared Awareness and Adaptation

A changed hierarchy does not operate by itself. This chapter examines the human-AI team that forms around a redesigned workflow: when an agent plans, explains, iterates, or acts toward a shared goal, how do people coordinate with it and keep their understanding aligned? The practical value is a way to judge whether role definitions, feedback, trust, and learning are strong enough to preserve human judgment as technical capacity moves between layers.

The structural model shows who executes, who validates, and who handles exceptions. Human-AI teaming adds a harder question: do the participants understand the same task and anticipate one another’s next move? A flatter reporting line can still contain confused responsibilities. A broader expert span can still fail if workers cannot tell when an AI answer is outside its reliable range. Conversely, a carefully designed team can use role changes to make the organization more adaptive without pretending that the agent carries human responsibility.

An AI becomes a team participant through interdependence

The distinction between a tool and a team member is functional. A passive tool waits for a person to activate it and returns a result that is treated as an instrument of the person’s work. An agentic system can plan a sequence, sustain interaction, reflect on an output, explain what it is doing, or prescribe and take actions. When humans and such an agent coordinate toward shared goals, plan and iterate together, and evaluate the result, the interaction has the structure of human-AI teaming.

This definition does not require treating an AI system as a person. The agent does not acquire human motives, rights, or final accountability merely because it participates in task execution. “Team member” names a role in a socio-technical system: the system can make choices that affect the work of other participants, and its behavior must be included in the team’s coordination model.

That functional distinction matters for the hierarchy question. A tool can be added to an existing role with little change to the distribution of authority. An active agent can alter task order, suggest a different allocation, identify an exception, or produce an output that another participant must evaluate. The team then needs explicit expectations about what the agent may do, what people may override, and how a change in one task affects the rest of the workflow.

The first coordination problem is shared situation awareness. Individual situation awareness has three connected levels. Perception is noticing relevant information in the environment. Comprehension is understanding what that information means and how a teammate’s contribution fits the shared task. Projection is anticipating what a teammate will do next and what consequences may follow. In a human-AI team, both the human and the AI need some representation of the task state, while the human team as a whole needs enough common understanding to act coherently.

Shared awareness is more demanding than giving every participant the same data. It concerns whether participants possess the same awareness of the requirements that matter for the joint task. A worker may see an AI recommendation, an expert may see the underlying exception, and the agent may be tracking a different objective inferred from the prompt. If those perspectives are not reconciled, the team can be informed yet unsynchronized. A shared mental model connects the observations to the goal, the roles, the constraints, and the likely next actions.

This gives role design a cognitive foundation.