Frederick Taylor came into the machine shop of the Midvale Steel Company in 1878 and started as a day laborer. He became the shop’s clerk, then ran a lathe. He turned out more work than the other machinists, and after several months he was made gang boss over the lathes.
The shop, by his account, “was really run by the workmen, and not by the bosses.” The men had “carefully planned just how fast each job should be done,” at about one-third of a good day’s work. As soon as Taylor took charge, one man after another came to see him. “Now, Fred we’re very glad to see that you’ve been made gang-boss. You know the game all right, and we’re sure that you’re not likely to be a piece-work hog.” If he tried breaking any of these rates, they told him, “we’ll throw you over the fence.”
He told them he was on the side of the management now. That, he wrote, “immediately started a war.” It lasted about three years. Output rose, in many cases doubled.
Then, looking back, Taylor made an admission. Although he was foreman, he wrote, “the combined knowledge and skill of the workmen who were under him was certainly ten times as great as his own.”
A desk and a lathe
Taylor’s conclusion was not to hand the work back to the men who understood it. It was to move the understanding. With the company president’s permission he began timing the work, and the system that grew out of it put planning in an office.
One group decides how the work is done; another does it as written.
All the planning the workman had done from experience, Taylor wrote in The Principles of Scientific Management in 1911, must now be done by management, partly because “it would be physically impossible for him to work at his machine and at a desk at the same time.” In most cases, he added, “one type of man is needed to plan ahead and an entirely different type to execute the work.” Each man’s work was to be planned at least a day ahead, and in most cases he would get written instructions saying what to do, how, and the exact time allowed.
The logic is easier to see in a kitchen. A recipe written by someone who tested it fifty times lets a cook on day two turn out the same dish as a cook in year ten. Nobody at the stove has to reinvent the sauce, and nobody at the stove gets to decide the sauce takes an hour.
At Midvale that was the point. The men’s plan had set a pace that, as Taylor saw it, kept them from being made to work harder for the same pay; he told friends who asked that in their place he’d fight it too. Moving the planning upstairs took the pace out of their hands.
The knowledge left behind
The split has had defenders. In a 1980 review, Edwin Locke of the University of Maryland argued that where planning is done expertly, dividing it from doing will likely raise productivity, though perhaps at some cost to morale.
Peter Drucker, in The Practice of Management in 1954, called the “divorce of planning from doing” the second blind spot of scientific management. It reflected, he wrote, “a dubious and dangerous philosophical concept of an elite.”
The sharpest test came from a car plant. NUMMI, a joint venture of General Motors and Toyota, began building cars in Fremont, California, in 1984, in a plant GM had closed. Its standards, once set, were to be respected down to the second. But the management researcher Paul Adler, in a 1992 study draft, found the plant had no industrial engineers responsible for defining methods, where the old GM plant had had 82. Teams set their own standards, which still had to match across shifts; by one manager’s description, workers timed each other with a stopwatch. Adler also recorded workers’ complaints about the system. The stopwatch was Taylor’s. The hands holding it weren’t.
The ticket that couldn’t say no
Imagine you lead a team of four developers and write every ticket yourself: what to change, which file, how many hours. Output is steady. Then a junior developer, halfway through a ticket, finds that the table she’s been told to add a column to is about to be retired. The ticket doesn’t ask what she thinks. She adds the column, logs four hours, and closes it.
You find out at the migration, two months later.
Planners and workers
In January 2026, Cursor described running hundreds of coding agents on a single project for weeks. Its first design gave agents equal status, coordinating through a shared file. In Cursor’s description, twenty agents slowed to the throughput of two or three, and with no hierarchy they “became risk-averse,” avoiding difficult tasks and making “small, safe changes.”
The fix was Taylor’s. Planners explore the codebase and create tasks. Workers, the post says, “don’t coordinate with other workers or worry about the big picture.” That solved most of the coordination problems, by Cursor’s account, and a run aimed at building a web browser from scratch wrote over a million lines of code in close to a week.
The analogy has limits. Taylor said his men held back for a reason, to protect their rates; Cursor doesn’t say why its agents played safe. And its own lessons pull back from pure Taylor: too much structure “creates fragility,” an integrator role was dropped as a bottleneck, and the system still needs fresh starts to fight drift. A large migration it describes “still needs careful review.”
If you split planning from doing, give the doer a way back. Let a worker agent reply that a task as written is wrong and why, then read a sample of those replies against the code. A worker that never sends one back may be doing what the junior developer did.