Work and Workflow

A bricklayer in 1890 laying a course of common brick stooped to the pile at his feet, picked up a brick, turned it over to find the better face, tapped it to knock off loose mortar, and carried it up to the wall. Each brick took eighteen separate motions. Frank Gilbreth, who had started as a bricklayer's apprentice in Boston and become a contractor, counted them. He then built an adjustable scaffold that kept the pile at waist height, had laborers stack bricks face up on a packet before they reached the mason, and changed the mortar so a brick could be pressed into place without tapping. Eighteen motions became four and a half. A man who laid about a hundred and twenty bricks an hour now laid around three hundred and fifty.

Gilbreth and his wife Lillian went on to break all manual work into seventeen elementary motions they called therbligs, their own name spelled nearly backward: search, select, grasp, transport, hold, position, release, and so on, down to "unavoidable delay" and "rest." Any job, filmed against a gridded backdrop with a clock in frame, could be written out as a string of them. What the Gilbreths saw is true of all work, whether or not anyone films it. From the outside work looks continuous. Bricks rise, a wall appears, a page fills. The word workflow carries the same picture, work moving like water through a channel. From the inside it is nothing of the kind. It is a sequence of discrete choices, each one a fork: this brick or that one, tap or press, which face out, stop now or lay one more course. A mason doing it for twenty years no longer notices most of the forks, but they are still there, and each was once decided.

Talk about freedom and work usually stays with the work: which job, for whom, at what wage, whether a person may quit. Those questions matter, but they are asked once, at the door.

The forks come thousands of times a day, after the door has closed. A person can choose a trade freely and then hold none of the choices inside it, and a person in work they never chose can hold most of them. A conscript cook who decides the menu has more say over a day than a surgeon whose every step is set by protocol. Liberty in working life attaches less to the work than to the workflow.

Who holds the forks

Motion study did not invent the forks, it moved them. In the older trade the mason decided how to lay a brick, and the decision was part of what made him a mason rather than a laborer. No contract said so. It was the custom of the trade, understood by masons, contractors and clients alike, and a custom like that is held in common or not at all. Gilbreth's method took each fork, decided it once at a desk, and handed the answer back to the man at the wall as a rule. Frederick Taylor said it plainly: the planning department would do the thinking. The bricks went up faster. The mason kept his trowel and lost his choices.

Nothing about that transfer is forced by the work itself. The same scaffold could have been a mason's own invention, offered to other masons and taken up by whoever found it useful. Gilbreth himself first worked it out as a mason. The difference between a better method and an imposed one isn't in the method, it is in who decides at each fork whether to use it.

How a worker is paid has a lot to do with where the forks end up. A man paid by the piece, a thousand bricks laid true, sells a result. Whoever buys the wall has little reason to care whether he taps or presses, starts at the corner or the middle, and the forks stay with him. A man paid by the hour sells his time, and once time is bought, the buyer has every reason to care how each minute is spent.

Supervision of motions follows the hourly wage closely. How View of Work Shapes Law and Land traced how the firm came to answer for its workers' private lives. Motion study is the same drift at a smaller scale, the firm answering for each movement of a hand.

A loom that stops itself

Sakichi Toyoda, a carpenter's son from a village south of Nagoya, built looms. In 1924 his automatic loom could change shuttles without stopping, and it had a device that halted the machine the moment a single warp thread broke. Until then a weaver had to watch each loom for a break, since a loom running on with a broken thread wove yards of flawed cloth. With the stop, one weaver could tend many looms, and no one needed to watch any of them continuously.

His son's company carried the idea onto the assembly line. Above each station at Toyota hangs a cord, the andon, and any worker who sees a defect can pull it and stop the line.

Of all the forks in an assembly job, that one, whether this work is good enough to pass, was placed with the person doing it rather than with an inspector at the end. Visitors from Detroit in the 1980s often found the cord hard to believe. A line that any one worker could halt seemed to them a line that would never run.

In 1984 Toyota and General Motors reopened GM's shuttered plant in Fremont as a joint venture, NUMMI, and rehired much of the same union workforce that had made it one of GM's worst. The cords went up over every station. Workers pulled them, a team leader came to help rather than to blame, and within a couple of years the plant was building cars as well as Toyota did in Japan.

GM later hung the same cords in its own plants, where they mostly went unpulled. Workers there had learned that a stopped line brought a supervisor's anger instead of his help. The cord was the same piece of rope in both places. The freedom it granted was written into the procedure in both, and in only one of them did it exist, because only there did everyone understand, without its being said, that the person who stopped the line had done the line a service. No worker could hold that understanding alone. It belonged to the team, the supervisor and the plant, and the written permission was the smallest part of it.

Toyota's line is no workshop of free craftsmen. Most of its forks are decided by engineers in exactly the way Gilbreth would have recognized. What the cord shows is that the forks can be sorted one by one, and a choice that seems to belong naturally to management can sit with a worker without the line falling apart.

Forks in a page

Writing has always been made of forks: which sentence first, which word, this example or that one, cut or keep. Until recently few of them were visible as choices, because there was one way to make a sentence, and it was to make it. Now nearly every step can be done two ways. A draft can be written by hand or generated and edited. A source can be read or summarized. A paragraph can be cut by the writer or tightened by a model. Each step has become a discrete choice in plain sight, and the work is no longer a single act but a sequence of them, each a fork between hand and machine.

Forms of Luddism described a writer who refuses the machine entirely and a church district that takes up tools one at a time.

Both are answers to the same chain of forks. One writer decides them all the same way, once. The Amish decide each separately, after watching what it does. Most people working now sit somewhere between, using a machine for one step and refusing it for the next, and changing their minds as they go.

Here too, liberty is a matter of workflow, not of work. A report is a report whichever way it was made, and the reader may never know the difference. The question is the one the bricklayer faced.

Where an employer writes a policy that every report goes through a model first, or that no draft may, the forks have moved back to the planning department. Where a writer chooses, step by step, which parts are theirs and which they hand off, the workflow belongs to them, whether the result was made mostly by hand or mostly by machine. Gilbreth's scaffold and a language model are alike in this respect. Either can be a tool a worker picks up or a rule a worker is handed.

Who sets the defaults

What a model changes is less the work than its grain. Gilbreth needed a film camera, a gridded wall and a clock to see the forks in laying a brick. A language model makes each fork in a page visible by offering to take it, and it offers to take more than motions. Gilbreth's planning department decided how a hand should move. A model decides, unless someone stops it, which point comes first, which example carries it, which word ends the sentence. Those choices arrive as defaults, and a default is a fork decided somewhere else, by people who never saw this page.

The mason after twenty years no longer noticed his forks, but each had once been his to decide. A writer who accepts a model's defaults can stop noticing forks that were never his at all.

Order matters as much as hand or machine. A translator in Lisbon working on a Brazilian novel can read the chapter first and then open the machine's draft, or open the draft and correct it. The two workflows produce pages that look alike. In the first she meets the author's sentence before the machine's, and the machine's version becomes a second opinion. In the second the machine's sentence is the one she meets first, and her judgment works at its margins. Same tools, same steps, and the forks sit in a different place because the sequence changed.

The cases already here offer bearings, though none of them is a rulebook. Toyoda's loom ran on its own so that one weaver could tend many, and the stop was what made that possible: a machine doing the work and a person able to halt it at the first broken thread. The cord at Fremont put the judgment of whether work was good enough with the person doing it. The Amish take up a tool one step at a time, after watching what it does to the life around it.

Piece work shows where the forks end up when a buyer pays for the wall and not the motions. An editor who asks for a finished report and leaves its steps to the writer keeps the forks with the writer. One who prescribes the steps, a model for every draft or for none, has rebuilt the planning department.

None of it holds in a workplace where choosing invites suspicion. If a writer who drafted by hand is thought slow, and one who used a model is thought a cheat, the choice is free on paper and gone in practice, the GM cord again. The direction AI takes in working life is being set now, less by policies than by what colleagues come to assume about one another: whether the person at the page is trusted to decide where the machine belongs.

Discrete and discretion are the same Latin word, discernere, to separate, to tell apart. A workflow seen as discrete steps is a workflow in which each step calls for discretion, and the old name for a worker who held that discretion was a master of the trade, a standing other masters conferred and no one could claim alone.

---

A note on how this piece was written: the idea of workflow as a chain of discrete choices, and the question of who holds them, are mine, as is the choice of examples. The sentences were drafted with AI assistance from that direction, then edited by hand, which is one of the forks this essay describes.