Part of a series:
This series explores the question, “can a machine do a human’s job?” We’ll start with examining what the term “work” entails.
We will consider the commercial enterprise. In broad strokes, a commercial enterprise functions by transforming inputs into outputs of value. This transformation generally involves the participation of labor (work) and capital, in addition to material or informational inputs. (Financial) capital commands the construction or employment of the machinery that aids the transformation, and labor applies human knowledge and skill to the process.
Any non-trivial enterprise is extremely complex, creating various kinds of output at various times and having a complex and changing web of internal operations, but to aid in our analysis we will consider the production of a single kind of output in a process that progressively accumulates additions and changes resulting in the final product. This unit on analysis is called a “value stream.” To give a few examples,
Along the length of the value stream, various inputs enter it. These may be untransformed (raw) materials. They might be the outputs of other value streams, such as a component produced by a separate manufacturing line, for example. Inputs might be informational, such as the predictions of a weather forecast, the locations of vehicles in a shipping fleet, or the measured length of a piece of steel. Whatever form they take, the inputs are incorporated into the product as it proceeds through the value stream by some kind of processing.
Processing transforms the material or information flowing through a value stream by incorporating inputs into it. Processing is often thought of as progressing in discrete steps, but it need not be so. Consider a plant continuously growing in a farmer’s field or beer fermenting in a brewery. However, to simplify the analysis we will proceed by considering “process steps.” This maps to continuous processes without too much difficulty—the activities of a farm on a given day in the growing season can be regarded as a step, incorporating the day’s sunlight and water as well as other mechanical or chemical interventions of the farmer.
A process step can occur either with or without “discretion,” defined as the freedom to decide what should be done in the particular situation. The situation is the specific circumstances and context in which the step takes place. In many situations, the “correct” decision is not well-defined, unique, nor can even be ascertained in retrospect.
Consider a step that involves a task in which no discretion is required: it is to press a certain button on the hour, every hour. No decision or judgement is to be made—the button is only to be pressed at HH:00:00 of the clock every hour (for some clock present at the processing step). Assume that some variation is allowed: the button press must occur ±X seconds of the actual top of the hour. For any value of X, this is a task that is perfectly “automatable.” There is no place in which discretion plays any part, so there need not be any human involved. A machine may be constructed to press the button to any degree of precision (smallness of X) that technology allows. A human could be tasked with the button pressing, depending on their skill, but this task is eminently automatable.
In contrast, consider craft work. A craftperson is an individual who has gained a lot of experience working in a specific domain. From a woodworker choosing a suitable piece of lumber for a project to a programmer modeling a problem domain in code, discretion is deeply integrated into the work. There are more or less correct choices, but they are also often difficult to ascertain until well after the fact and often only under failure conditions, e.g. a chair breaking prematurely or a difficult code refactor. Both because of this long feedback loop (if it ever closes) and domain complexity, the worker’s discretion guided by their deep experience is heavily relied upon.
Let’s consider another kind of processing step. It is to create a design for a promotional marketing poster for a new product that will be introduced. This task requires the consideration of a wide ranging amount of context: target market, product attributes, aesthetics, and so on. In realistic scenarios, the designer faces many constraints such as brand, but in most cases much discretion is left to the designer. She is a craftperson, relying on her experience and skill, but in this case we emphasize the open-endedness of the context to be considered for the task. The result may be much more ambiguous as well. Following a marketing campaign, it is not always clear what contributed to the failure or success of the product launch, and to what degree.
We have been a bit inprecise in their use of the word “discretion,” so let’s parse it. Our examples describe discretion that relies on the experience of the worker, and this is one important facet. Another is in our initial definition: that it is the freedom to decide. This freedom implies a moral dimension—discretion is inseparable from accountability.
The steps in a value stream incorporate inputs into the product moving through the stream, changing it. Generally, a process step is “variable” in its performance meaning that, given identical inputs entering the step multiple times, the output will not be the same each time. In the button pressing task, the performance of the step is such that the button is pressed at HH:00:00±Y of the hour. Presumably this affects whatever the product is that is processed by this step, injecting randomness. Note that the performance variability is different from X given in the task specification. It has been determined that as long as Y <= X the step can be considered successfully performed. The value X is the “tolerance” of the task, or the amount of variability that can be tolerated.
In automatable processing steps, performance variability must be bounded. Sometimes it is quantifiable as a numerical tolerance but often not. Variability must be bounded to such an extent that it does not qualitatively or quantitatively change the value of the product at the output of the value stream, effectively changing it into a different product.
Usually, much effort is spent driving variability out of a value stream. This is an effort to create a product of consistent quality. Sometimes, however, variability is impossible to remove, and this introduces a sorting or quality control step in many value streams. Defective product is scrapped (discarded) or, as some enterprises have creatively figured out, diverted into varying product grades. An example here is performance grades in microchip manufacturing. Another is sorting carrots into nicely straight, tapered specimens fit for the grocery shelf and weird carrots that will be turned into “baby carrots.”
Non-automatable processing steps are those whose performance is imbued with discretion. At these points, judgement is substituted for specification, and the variability of the step is essentially unbounded. The step may involve dealing with the unexpected, taking into account factors that are not readily apparent or present in the product or immediate context, or a creative leap. It might involve a seemingly arbitrary choice among alternatives that only experience may inform, but which have real and potentially unforseen consequences on the product. Discretion injects variability into the value stream, and its consequence may be success or failure, depending on the skill and experience (and sometimes luck) of the worker. (I was introduced to this concept of “performance variability” in the excellent talk by Ryan Kitchens, an SRE at Netflix, How Did Things Go Right? Learning More from Incidents)
When a processing step is automated, a “machine” performs it. We are vague about our definition of machine here. It might be an oven, steel roller, coffee grinder, or a computer program. It doesn’t matter much. The key fact is that the machine is the output of a separate value stream that feeds into the one employing the automation. Everything discussed about value streams here also applies to this one.
The above discussions of discretion involved that which takes place within a value stream processing step and contributes to the performance of the processing step. We’ll call this “work discretion.” In contrast, we call the ability to perform an automatable processing step to specification and within tolerances “work performance.” One effect of work discretion is to add variability necessary for the value stream to remain robust in the face of an uncertain world.
There is another type of discretion generally seen in enterprises. This is “managerial discretion.” This type of discretion operates on the value stream, deciding when and what the stream should produce and what human or automated resources shall be used. Managerial discretion of course also implies accountability for the decisions made. As noted above, accountability is present in all places where discretion is exercised, but enterprises are typically set up to formally centralize accountability in its management structure.
In large enterprises managerial structures have multiple layers and form a hierarchy. Front-line managers who interact directly with value streams are accountable for the outputs and costs of the stream that they oversee. Higher level “managers of managers” are accountable for the costs and outputs, in aggregate, of all the value streams under them, and their decisions affect how these streams interact with one another. This hierarchy continues upwards to the highest level of management that is accountable for the entire enterprise.
A job comprises the activities that a person does during their employment by an enterprise. Depending on the role, the composition of the job will be a mixture, in varying proportions, of work performance, work discretion, and managerial discretion. When performing work within a value stream, work performance and discretion will be applied. Multiple value streams will be engaged with throughout a job in most cases, and managerial discretion decides when, which, and how the value streams will be worked on. Sometimes the job holder makes these choices, applying managerial discretion themselves, and sometimes they will be directed to their task by another whose job entails such discretion.
The composition of these various components into a job for a given human gives that person financial resources to exist in society. Depending on the compensation, a person may be adequately served by one job, or they might have to cobble together a living through multiple jobs. A job may be long-term, or it might exist for only a limited time, causing the person to have to continually seek further employment.
We will more deeply explore the design of work, and consequently of jobs, in the next part after we have developed tools for examining them through a financial lens. For now, it suffices to say that the way in which jobs are designed has serious implications to the ability of humans to exist stably in society.
With this framework in place we are now ready to discuss how work is designed in more detail, which we will do in the next part of this series.