A tracker keeps flagging the same four minutes at an airport stand. Baggage carts cross the catering lane on nearly every arrival. A supervisor walks the stand between turns and shifts the staging point, reclaiming the lost time. Chapter 7 offers this as an imagined example, rather than an account of a particular airport trial.
That promise is hidden from the passenger but carried by the airline. The passenger sees a boarding time, a destination, a seat number, perhaps a connection tight enough to make the next gate feel like a test.
The ramp is a clock with weather inside it: equipment, people, arrivals, and plans shift together. Chapter 7 sets its illustration alongside discussion of humanoids entering airport operations, then describes a tracker noticing the repeated crossing. The tracker helps expose where time goes; the supervisor has to work out what the measurement means.
A task is not the whole work
Moving a bag from one place to another can be specified and timed. The airport operation around that movement keeps changing as people act. A delayed arrival shifts the turn. A cart parked elsewhere narrows a route. A worker steps into a machine’s path. The right response depends on what else is happening and what the delay might put at risk.
That difference matters because complicated and complex describe different problems. A humanoid is complicated: engineers can examine its parts and tune its movement, then test whether its sensors respond as intended. The work is complex when people must read a situation that changes as they respond to it. An excellent test of the machine cannot, by itself, settle how it should fit into that work.
The four-minute example is easy to overlook. The tracker can show where time goes; it cannot decide that a cart’s staging point is the cause, or whether moving it is safe during a busy turn. The supervisor leaves the dashboard, walks the stand, and sees how the crews and equipment share the space. A modest operating adjustment returns time without asking anyone to move faster.
The supervisor’s modest adjustment returns time; retuning the humanoid to move more quickly might also affect the turn, yet it leaves the staging question unanswered. The organization has to know which problem it is trying to solve before a good number becomes its answer.
What does the machine’s pause change?
A safe pause can still change the operation. Once the humanoid stops, a worker may need to decide whether it detected a hazard, lost its route, or is waiting for an instruction. The next cart may be held back. The supervisor is still watching the turn clock. The machine has obeyed its limit; the people nearby now have another decision to make.
This is where an assignment such as “move baggage” leaves out too much. A usable description has to account for the right bag, the changing route, the people in it, and the airline’s promise to the traveler. If the measure is only cycle time, a stop may look like lost efficiency. A separate safety measure may treat a machine that waits indefinitely as a success. The work requires someone to see both what the system is doing and what its action is doing to the rest of the operation.
The same distinction appears far from an airport ramp. An AI system can prepare a summary, rank a case, or draft a reply as a clearly bounded task. The surrounding work may still depend on a person noticing which fact changes the meaning, or what the recipient will reasonably rely on. A clean output does not guarantee the situation has been understood. The shape of that work is less visible in an office, but the need to interpret the effects remains.
Organizations often decide who sponsors a machine before they decide who owns its effects across the work. A named sponsor can answer questions about performance. A team can track errors and adjust settings. Yet a pause, a recommendation, or a queue change may affect another group whose people now have to absorb the interruption. A box on an org chart cannot say which person may hold the process while those consequences are understood.
Give someone room to read the situation
Before a new system enters live operations, leaders can specify the conditions under which it should continue or stop, then name who can respond when those conditions appear. The person closest to the work needs a reliable way to halt a routine sequence without being marked down for slowing it. A supervisor with wider view needs time to assess what changed and authority to decide how the work resumes.
Those permissions matter only if the operating measures recognize their use. If every safe interruption counts against a crew’s turnaround target, workers learn what the organization actually values. If a human review is recorded as approval while the reviewer has no practical way to change the recommendation, the recorded check tells leaders little about what judgment occurred. The design has to account for moments when the plan no longer fits alongside routine completion.
This will add pauses in some places. It may also reveal a physical bottleneck that a more capable machine would have worked around without fixing. The four-minute illustration does not claim that every delay has such an easy answer. Its value is smaller: someone went to the place where the work happened and asked what the number left out.
At the stand, the turn clock keeps running while the supervisor studies the carts. A system can help bring that repeated crossing into view. It cannot decide whether a staging point should change, or who will stand behind the interruption when the next departure is already late. The crews are waiting for the route to clear.
This essay draws on Chapter 7, “The Machine Is Complicated. The Work Is Complex,” of AI in the Org Chart.