08. When Does Efficiency Become Violence?
The Problem: Smaller Numbers and Longer Nights
A number worth celebrating appears on the logistics center’s dashboard: order-processing time down 18 percent, picking routes shortened by 23 percent, a record number of items processed per hour. There is applause in the meeting room. An algorithm calculates the next item each worker should pick and the route they should take, while the wrist terminal vibrates without pause.
That night, the site manager checks the workers’ break records. Breaks have become shorter, and the time it took to return after a trip to the bathroom has been marked as a drop in productivity. The system was not designed to punish anyone. It merely calculated delay as a cost. But the more finely tuned the calculation becomes, the more human bodies begin to be treated as sources of system error.
Efficiency is a good word. It reduces waste and waiting, making it possible to do more with the same resources. The problem lies in what efficiency takes as its standard. Saying that time has been reduced hides whose time was reduced. Saying that costs have been cut erases who paid those costs with their bodies and their anxiety.
The Philosophical Question: When the Optimization of Means Swallows the Ends
Efficiency is originally a means to an end. Trouble begins when a means that is easy to measure takes the place of the goal. Throughput, click-through rate, response time, and model accuracy are easy to express as numbers. Dignity, autonomy, trust, and time to recover are much harder to see on a dashboard. When we treat what cannot be seen as unimportant, the system improves only what is visible.
Philosophy turns the question of optimization around. Instead of asking how we can process something faster, it asks: What should we refuse to process quickly? Not every delay is waste. The time a support agent spends listening to a user all the way through, the time a doctor spends checking an unusual symptom once more, and the time a developer spends following a troubling line in a log may look inefficient on the surface. They are buffer zones that prevent accidents.
Automated systems like averages. They create rules around the average user, the average work pace, and the average level of risk. But life breaks down not in the average, but in the deviations. A pregnant worker, a customer who is not fluent in the language, a user with an old device, or someone whose livelihood can be shaken by a single mistake pays a greater cost simply because they fall outside the average.
An Engineering Scene: The Call Center’s Thirty-Second Rule
A call center decides to automate its evaluation of customer-service agents. After each call, the system analyzes its duration and scores periods of silence and the phrases used by the agent. The goal is clear: reduce average call time and handle more calls. At first, the results look good. The queue gets shorter, and the reports repeatedly use the phrase “improved efficiency.”
But agents begin transferring difficult customers elsewhere. A customer with hearing loss, an older person who cannot understand the refund policy, or someone who has already heard the same explanation several times will inevitably require a longer call. The agents want to help them properly, but staying on the line too long lowers their individual scores. In the end, the system makes it a rational choice to cut off the people who need the most help as quickly as possible.
When the team discovers the problem, its first thought is to retrain the model. But predictive performance is not the core issue. The problem is how good service was defined. The team proposes looking at first-call resolution and repeat-contact rates alongside call duration, and creates fields where agents can record exceptional circumstances as well as a process for challenging their evaluations. Adding more numbers will not solve every problem. But it can at least change the structure so that one number does not govern a person’s entire behavior.
Counterargument and Tension: Do Not Romanticize Inefficiency
Criticism of efficiency can sometimes fall into the trap on the opposite side: the romantic idea that slowness is inherently humane and that avoiding automation is inherently safe. In reality, wasted resources put pressure on someone’s life. Long waits at hospital registration, repeatedly entering the same information into administrative forms, and complicated procedures inaccessible to people with disabilities are already forms of violence against particular people. Technologies that improve efficiency can reduce those burdens.
The issue, then, is not a choice between efficiency and humanity. It is a question of whose perspective defines efficiency, and with whom its results are shared. A good system does not merely increase throughput. It leaves users room to explain their circumstances, makes it possible to treat exceptions as exceptions, and brings unmeasured losses into view. It offers a faster path without turning that path into a compulsion.
An engineer performs a political act the moment they decide what to put into an optimization function. A single weight, timeout, or retry count can change someone’s experience. So next to every performance metric, there should be a question: After this number went down, did someone’s night become longer? What invisible labor made the rise in this graph possible?
A Question to End With
What does your system classify as waste? Does it treat waiting and hesitation, explanation and care, inspection and rest as nothing more than friction to be removed? We need to ask again whether building a faster system and building a system people can live with are really pointing in the same direction.