Part of a series:
This series explores the question, “can a machine do a human’s job?” In the first part we laid out a framework analyzing various kinds of activity involved in work, as well as the concept of automation. We touched briefly on composing the different kinds of work into jobs.
In this part, we will dig more deeply into how value streams are designed and composed in an enterprise, developing some financial concepts with which to consider the topic. We will speak more about how jobs are designed, bringing in some historical context through management philosophies that shape jobs up to the present day.
An enterprise must decide what is to be produced and the work performance specifications that will apply to its processing steps. These decisions comprise the activities of managerial discretion. In many organizations this work is performed by specific value streams set up within the enterprise. To state a few examples,
The work performed by these functions is heavily imbued with discretion as individuals applying it navigate ambiguous contexts and make judgement calls based on their experience. Although these activities do not directly generate revenue for the enterprise, it’s incorrect to say that they do not create value at all. The value streams they participate in have as their outputs value that is entirely internal to the enterprise.
In small or less mature organizations, the functions described above may be performed by less stable value streams, perhaps in ad hoc decisions by the owners. Nevertheless, they can be described as value streams, producing value for the enterprise, and whose steps are mostly or entirely discretionary.
In talking about commercial enterprises, the flow of money determines much of how things are set up. The product at the output of a value stream is valuable to someone, and so brings money into the stream. If product, material, or information flows down the value stream to its output, then money flows up it. At each processing step, money is apportioned to the inputs at that step. This could be paying a human worker for their performance of the step, paying a machine for its automation of the step (equivalent to paying the value stream that produces the machine), paying a supplier for raw materials, paying rent on land used, or paying for intermediate product—the output of the previous processing step. Money is also apportioned to the act of exercising work discretion, so a worker’s experience and accountability commands a price.
Managerial discretion must also be allocated an amount, but may be difficult to assign to any particular processing step. To simplify the analysis, we will say that the above stepwise apportionment continues upstream until all inputs are accounted for. Following this, money that is leftover is allocated to managerial discretion and return on capital. It’s perhaps a matter of negotiation between investors and representatives of the management hierarchy to decide this split.
This picture is quite idealized. In practice, it is very difficult to precisely allocate the money flowing up the value stream. In addition, the enterprise evolves and changes over time, so any configuration of value streams and financial allocations will be unstable and invalidated with new information flowing into the enterprise. Financial allocations must, in general, be given a degree of slop or wiggle room to absorb these changes.
Given any snapshot in time, however, financial considerations inform the decisions made about how value streams are set up. We will consider how for automatable and non-automatable processing steps in turn.
Setting up an automatable processing step involves different things depending on whether a human or machine will do the work. When a machine is performing the step, it involves setting up an entirely new value stream to produce and maintain the machine. Sometimes this setup occurs (or already exists) within a different enterprise, and the machine is purchased or rented. Alternately, the enterprise may set up the value stream within itself, avoiding payment to a separate enterprise, and also building its own internal capability. The choice of whether to automate (have a machine do it) or not (have a human do it) depends on several factors.
The first and most obvious factor is called startup costs. In the case of a human worker, startup cost is their training for the task. For a machine, it is purchasing or renting the machine from another enterprise or setting up a value stream to build and maintain it. And it is typically more costly to start up a worker or machine from rest or another task than it is to continue running an already “warm” performer. Maintenance costs are those costs incurred to ensure the processing step continues to operate at the specified level of performance over time. For humans, it might be ongoing training and skills growth. For machines, it is continuing to pay the value stream that produces the machine or another stream that provides for its continued operation.
The choice of which of these costs to pay depends on circumstances specific to the context in which the work is being done. Is the work a one-time or occasional thing, not repeated often or at all? If so, then it might make more sense to pay a human to perform the work rather than paying the costs of automating it.
The available technology has a lot to say about whether a machine can be used to perform a processing step. In an automatable processing step, the specification or tolerance determines what technology must be used to create a machine that can perform the task. The smaller the tolerance, the more sophisticated the required technology, and, generally, the more complex and costly the value stream that produces the machine must be.
A human might be able to perform an automatable processing step to tight tolerance, sometimes even greater than a machine can for any available technology. I’m reminded of a remarkable man named Achim Leistner who hand-polished the world’s roundest object. But there are even more pedestrian tasks at which humans excel and machines struggle, such as operating an automobile in the presence of uncertain road conditions, other vehicles, and, uh, pedestrians (although advances are apparently being made in this area, I’m compelled to use this example for the pedestrian joke).
In many cases, there is a choice between human or machine performance. There is a limit to the amount of money that can be allocated to work performance at a processing step, and this will strongly inform the decision.
We will now consider the lifecycle of value streams. In some enterprises value streams are large, visible, and long-lived. These types of streams probably have large associated costs and therefore can only be used for high-valued outputs. Other value streams might be ephemeral, only existing briefly to create an output and then disappearing.
In our discussion above, we considered a single value stream producing a single kind of output. This description is simplified. There might be multiple varieties of the final product, which entail adjustments to the stream’s configuration. There may also be changeover of the product or rapidly switching amongst products. The decisions that choose these configurations may be made by a formal manager or by the workers themselves.
Each time a value stream changeover or reconfiguration occurs, there are generally startup costs associated with it above that of the steady-state cost of operating in one configuration continuously. For human workers, there are cognitive context-switching costs and perhaps training. For machines there are startup costs or reconfiguration costs of the value stream which produces the machine.
The decision to use machines or humans depends on these costs, technology, and the real or expected lifecycles of the value stream(s) involved.
A non-automatable processing step is basically one in which work discretion cannot be removed from the performance of the task. The discretion is ingrained in the nature of the task itself or is a consequence of the material or context in which the task is to be performed. The task may involve rapidly switching between applying work performance and work discretion, and the work performance phase of the task may be aided by “tooling” which is a type of machinery (again, created and maintained by some separate value stream). Some examples from various crafts can help to illustrate.
A woodworker might use hand tools—knives, chisels, and the like—to process his work material, wood, into a product. There are several levels of discretion that are applied. The first is managerial discretion deciding and planning what to produce in the large scale, which we will ignore for the moment. At a smaller scale, the worker makes many small decisions in deciding and adjusting details in the cuts and joins of the wood. At an even smaller scale, when passing their tooling through the material, he makes innumerable micro “decisions” about the application of force as the tool passes through unexpected variations in wood density and grain.
There are various ways that the discretion may be removed from some levels of the process. One might be the substitution of powered tools for hand tools. When this is done, then the micro “decisions” are less relevant as the tool can power through the variations with ease. This can end in failure, however, if there is excessive chipping of the wood as it is being worked. This failure might be overcome by guaranteeing more consistency in the wood input to the step and more precise application of the tool.
Slowly, as we remove variation in material and processing, and apply more energy and power to the process, we can see a heavily automated manufacturing line begin to take form. However, this entails a transformation of the product: from a hand-crafted one to a mass-produced one (mass-production being necessary to justify the cost of all this automation). Evidently, the discretion applied by the hand woodworker is a part of the appeal and the value of a hand-crafted wood product.
Programming involves the creation of an information processing machine. In most cases this machine performs data manipulation for another value stream—is an automation—but, as we saw before, automation is itself the output of a value stream, so we will consider the “processing step” of programming here.
The programmer takes as her inputs the requirements of the program to be written (an output of the “product discovery” process step), the architecture that she is contributing to, the software interfaces that it is to interact with, algorithms (known or looked up using Google, Stack Overflow, LLM, etc.), and the machine that it will run on. Her tooling comprises a text editor or integrated development environment, a test suite (that she is expected to write herself, generally), the compiler, etc.
As the programmer writes the program she has to constantly make many small but consequential decisions. How should this function be named? What interface shall be publicly exposed? What is the most efficient way to execute this process? And so on. Her tooling can help by providing things like auto-completion, but she exercises discretion in deciding whether or not to use what is suggested. Her queries using a search engine also involve the judgement of whether what is found is applicable or can be adapted to her current problem. She considers factors involved in later processing steps. Will this name be confusing to reviewers and future developers? Will this interface leak implementation details and become a maintenance nightmare? Does the design conform to architectural principles set forth by the team?
The questions the programmer must consider constantly throughout her work are frequently open ended, and the decisions she makes must take into account various social and technical contexts.
There’s no decision between human and machine performance here. Non-automatable processing steps, by definition, cannot be automated by a machine unless it is able to apply work discretion, and therefore be held accountable for its decisions. This seems like a silly notion, but we’ll look at it more carefully later on. The actual decision to be made is whether to replace the non-automatable process itself with an automatable one. Sometimes this is impossible because it changes the nature of the product, as in the case of hand-crafted wood products. Sometimes it is technically impossible or infeasible.
As we discussed earlier, at a processing step money will accrue to the work discretion as well as work performance. The presence or absence of work discretion therefore influences the cost of the processing step, so the removal of it (thereby changing the step to an automatable one) would have the effect of eliminating that component of the cost. This may appear worthwhile if it does not result in increased costs that wipe out the associated savings, or if it decreases the processing time of the step such that the revenue flowing into the value stream can be increased.
An often overlooked consequence of such a decision is the reduced flexibility of the value stream to adapt to uncertain or changing conditions since, as we discussed previously, the performance variability injected by work discretion helps to increase its resilience. For a fun fictional example, in Vernor Vinge’s A Deepness in the Sky (one of my favorite books) the pathologies of automation were frequently so great as to wreck entire planetary societies when their systems, for example, experienced a “deadlock.” My interpretation is that the author is not speaking strictly of computerized or mechanized automation. A political system can employ “automation” when choices are narrowed by strict procedure and rules. Legal systems remove discretion (employ “automation”) through deference to precedence and statute.
It is worth noting that automation can be designed in such away that discretion can be injected at points when it is needed. A real-world example comes from the products of Toyoda Automatic Loom Works (whose ideas were later built upon by Toyota Motor Corporation) which created textiles automatically but that, when a string break was detected, halted to allow a human to intervene. This principle was called “autonomation” or “automation with a human touch.” In the software industry we frequently embody this spirit, designing automated pipelines that stop and send out alerts for human intervention in case of faults.
In commercial enterprises, management is preoccupied with profit and/or growth. The enterprise’s investors require either the distribution of profits or the reinvestment of those profits into growing the principal investment. The management structure is accountable for delivering on this mandate. Now, in a privately-held enterprise the investors may be satisfied with operating at a loss if other desires are satisfied, such as social or environmental good, but this is not common. In publicly-owned enterprises there is, at least in the United States and to a lesser extent elsewhere, a legal requirement, to which the highest officers of an enterprise are held accountable, to deliver the greatest possible financial return to shareholders.
For either goal, profit or growth—and even when operating at a loss—decreasing the costs incurred by the enterprise is a paramount concern. At any moment in time, a given amount of revenue is flowing in due to the sale of the its outputs. If the amount of money allocated to direct inputs in the value stream can be reduced, then there is more leftover for management discretion and investors (profit taken either as distributions or reinvested into the enterprise to drive growth).
There are a variety of ways to reduce costs in the value stream. More efficient use of material or automated resources is one example. In a great many cases, however, the costs allocated to the human workers in the value stream are the greatest component, so reducing the amount spent on humans is often the most effective way to reduce costs. This may be achieved in a couple of different ways.
One is to replace a human performing an automatable processing step with a machine. This may be a virtuous change if the work that was done was extremely manual and required little skill, a soul-draining task that we call “toil” in the software industry. If, on the other hand, a highly skilled worker who had spent a lot of time refining and improving her hand at this task were replaced by a machine, then the outcome might rightly be called tragic. In either case the cost of the machine, amortized over time, must be less than the cost of the human performing the task over the same period, or the processing time must be reduced to such a degree as to allow for decreased takt time (delivery time, and therefore, increased revenue) for the value stream. The opportunity cost for the worker performing the task may also be such that the alternative uses for her energies also create a financial justification for the change.
A second way is to replace a non-automatable task with an automatable one. We often hear this in the context of replacing “skilled” labor with “unskilled.” Presumably, the unskilled laborer has less developed skills and experience in the task at hand, and therefore is less able to exercise discretion in performing the work. We said before that work discretion commands a financial allocation, so removing that component can greatly reduce the cost of the processing step. It can be the case, however, that the skilled worker is not actually replaced. Perhaps she has nowhere to go and stays in her job. If she’s lucky, she can make alternative higher-valued uses of her energy and time, but if not, then eventually, unless her pay is reduced, she will eventually be seen as an excess cost to be cut.
The humans in the management hierarchy are generally of small number compared to the workers performing direct work within the enterprise value streams. As they perform tasks that are almost entirely discretionary, they are often seen as the “brain” of the enterprise, while the workers are seen as its “hands.” Since they are of small number, and performing what is considered high-value management discretion, they can command much higher financial allocations (on an individual basis, if not necessarily in aggregate) than the bulk of the workers in the value stream.
If you will suffer the metaphor a bit longer, a certain type of brain tends to think highly of itself as compared to hands and feet. It wishes to exercise complete control over the body, coordinating and orchestrating its motions. If a hand were to move independently this brain would panic and reassert control. If the right hand didn’t know what the left was doing, resulting in a fumble, the brain considers this its own failing or blames the hands for their inability to function as desired. The brain sees itself as the control center—the true being—of which the body is merely an extension.
If management tends to think of itself in this way (not to say that they consciously do #NotAllManagers), then it will view with hostility actions of workers that it cannot exercise total control over. Even if the worker in the value stream is “rowing in the same direction” as the ship of enterprise, all of his actions aligned with maximizing shareholder value, the possibility that he might hold a different goal in his heart gives management unease. The idea that he might make choices about the speed or technique that he applies to his work implies variability in his work performance. We’ve noted that automatable work is done to a specification, and that variability within tolerance is technically allowable, any deviation from nominal performance is seen as waste to be eliminated, regardless of specification. Further, skilled labor that is non-substitutable represents a risk if the worker chooses to end their employment (I realize we haven’t talked about risk, but at risk of making this article too long, I choose to skip the topic. It’s super important, but I can get my main points across without it).
Does management in today’s enterprise think of the worker in this way? It definitely must consider many of these things, but I will not cynically declare that managers necessarily feel the feelings described above. However, it is undeniable that the makeup of the commercial enterprise creates structural forces that compel management in this direction. The presence of accountability for profits and growth shapes the enterprise at all levels. Regardless of aspirations and mission statements, the commercial enterprise (especially in the corporate legal form) must always yield to profits through revenue maximization and cost minimization.
Removing work discretion serves management’s needs in two ways. First, as we described above, it removes cost from the value stream, directly impacting the profitability of the enterprise. Second, it supports the psychological need of humans to have control over that which can affect their destiny, negatively or positively. That is, it removes discretion and accountability from the value stream, concentrating it in the management hierarchy.
This drive toward concentration has a long history. By the late 1700s the proliferation of standardization and interchangeable parts in manufacturing was ushering in the industrial age. Rather than relying on custom, one-of-a-kind parts fashioned by skilled craftspeople, it began to be possible to mass-produce parts that could interchangeably be substituted for one another. This kind of production used standard work procedures and tooling and could be done by relatively unskilled labor, allowing for a larger labor pool and lower wages overall.
It reached its pinnacle in the concept of “scientific management,” also known as Taylorism due one of its pioneers, Frederick Winslow Taylor. This philosophy of management emphasizes such techniques as “time-motion studies,” in which a worker’s movements are examined and timed using a stopwatch, and all extraneous movements are to be eliminated. A worker’s motions are precisely prescribed, and the ultimate goal is to effectively remove all discretion from the work. Total concentration of discretion and creativity in management is the ultimate goal and aim of scientific management.
These management philosophies had various effects on society, many of them tragically dislocating for labor, but also increasing the material efficiency and profitability of industry. Much (mostly worker) blood has been spilled seeking the balance between these concerns, and battle is being fought in workplaces to the present day. Today, the management intellectual pendulum has swung back from the extremes of scientific management as it is widely acknowledged that some degree of intelligence and discretion simply must be left in the hands and brain of the worker to generate business agility, resilience, and innovation (oh, yeah, and profit). It is an economic fact that automation is often ill-suited to the needs of the enterprise, and that non-automatable work has an indispensable place. Yet, the structural forces remain.
The typical full-time employment arrangement has a human worker performing various kinds of tasks for an enterprise. A non-management employee will sometimes be preoccupied with the “trunk” of the enterprise, the value stream that brings revenue into the enterprise. They might be involved in one of its “branches,” an internal value stream that feeds into the trunk. Or they might sometimes even be performing managerial work. A typical job will have the worker engaged with a specific value stream most of their time on the job, but their day often comprises a mix of these activities. Sometimes the worker is a generalist that bounces all over the place in the enterprise.
Shifting a human from one task to another involves a cognitive shift, as well as a physical one in many cases. And the choice of if and when such shifts are to be made is managerial discretion. Both of these represent costs. The former pure waste to be eliminated if possible, and the latter one to be exercised by the management hierarchy if possible. Indeed, many workers would like nothing more than to be “just told what to do,” disliking the ambiguity and accountability involved with applying managerial discretion, and preferring to simply apply their skills in work performance and work discretion.
Thus, the interest of the worker and management are often harmonious with respect to how a job is composed. Management would like to avoid the cost involved with task switching, and the worker would like to avoid the accountability for managerial discretion, which management is tasked with and compensated for anyways. An effort is therefore made to keep a worker engaged uninterrupted in a single task for as long as possible.
There are peculiarities to the human psyche that influence their job. Skilled workers usually receive intrinsic reward for applying work performance and discretion. They seek a “flow” state of being focused and immersed in the activity. Of course, most humans also seek growth. They become bored with a task or, after achieving a sense of mastery, seek other domains to conquer. Some humans are naturally generalists and feel reward in the act of switching tasks. Fitting the right human into the right job and appropriately evolving their job over time is the essence of people management. Contrast this with the machine, which can perform the same task tirelessly over and over, demanding none of the costly job affordances of the human worker.
Management philosophies and the law have shaped the composition and definition of jobs in important ways. It is somewhat arbitrary to divide management from workers in the way that we have, to be honest. If the managerial discretion applied to the value stream is itself a value stream, then there should be little justification in a differing conceptual treatment. Indeed, some of the things like product management and engineering that we discussed above are, while highly discretionary and “managerial” in a sense, performed by “workers” and not “managers.” What accounts for the distinction, then?
One of the more important distinctions is the legal one between people managers, with their attendant power to hire and fire, and workers with no people “under” them. The legal system places great accountability, through labor regulations, on these managers. Another distinction is the officers of the enterprise, mentioned previously, who are held responsibility for its financial performance. And there are other factors, such as professional licensing, that create accountability structures for certain kinds of workers.
The legal peculiarities of accountability cause tension when discretion is formally separated from it. For very simple value streams there is usually good enough alignment. In large enterprises, when accountability is formally concentrated in the management hierarchy and discretion is distributed in the workforce, we see the tension play out as management tries to drive discretion out of the work. We’re starting to see a separation of formal accountability from discretion here. While we still maintain that moral accountability cannot be separated from discretion, formal accountability as allotted by human organizations may create an artificial separation.
Management choices can have potentially tragic consequences on the humans involved in work within the enterprise. Very often management is incentivised or required to take these very measures. In the name of economic efficiency we have seen skilled workers be replaced by unskilled ones supported by systems of automation.
A full-time job can be broken into pieces when the component parts are divided amongst contractors performing “gig-work.” The impetus for this effort is to drive out any wiggle room that the financial allocation to the work may have, as well as to increase the utilization of labor resources. Modern platforms have reduced the transaction costs associated with finding a worker to perform a one-off or ad-hoc task. Supposed benefits to the worker in these arrangements are “flexibility” and “unlimited money as long as you hustle,” but more often the worker gets nothing but precarity.
The use of contract labor can relax the tension caused by discretion and accountability misalignment, since the worker is made into an entrepreneur. Gig-work “entrepreneurs” are usually in name and legal status only, as the platform exerts so much control over all aspects of the work.
This post has illustrated some of the main factors that go into deciding what and how work will be done in the enterprise. We have introduced financial concepts describing how money flows into the enterprise and costs are allocated, how management incentives and mandates drive decisions about how the work is designed, and how these choices affect jobs. This gives a good base from which to discuss automation technologies such as “AI,” but before that we must interrogate some of the assumptions and assertions we’ve made about the moral concept of accountability. We have asserted that discretion always entails accountability. This claim demands further investigation, which we will do in the next part of this series. We will also look at the notion of formal vs. moral accountability more closely.