Did Numerical Control Deskill the Machinist? Anatomy of a Craft Transformed

The episode opens with the anxiety that spread through machine shops as numerical control took off in the 1960s: would this new automation reduce the master machinist to an unskilled button watcher? The host notes he touched on the subject before and now takes it up in full: did the rise of NC mean the fall of the old-school machinist?

First the craft itself is introduced. A machinist runs powered tools that cut, shave, or grind metal, a trade set apart from casting or forging. A good hand did more than cut: working with designers, reading two-dimensional drawings, picturing the part in three dimensions, and catching pitfalls early with geometry and trigonometry. Then there was feel: sensing a half-thousandth-of-an-inch difference by sound and vibration. An 1887 manual holds that nothing in lathe work demands as much judgment and watchfulness as feed and speed. The floor was dangerous, where one mistake could cost a finger, covered in chips, reeking of oil, and extremely loud. That harsh setting bred a distinct masculine shop culture.

In the first half of the twentieth century, machining was the largest skilled manufacturing trade for American men. The classic route was a four-year apprenticeship of about eight thousand hours, rising to five years and ten thousand for tool and die makers. A memoir of a 1940s General Electric apprenticeship describes basic shop training plus three nights a week of shop math, physics, blueprint reading, and electricity, followed by long rotations across machines. At the end the company certified the newcomer as a journeyman, with a tool kit or bonus and a raise. Yet this was not the only road: self-taught hands raised on shop experience and home study existed, alongside courses claiming to teach boys the trade, which the host likens to the coding boot camps of the 1930s. Some workers never became full machinists and stayed on as semi-skilled single-machine operators. Shops ranged from small-run job shops to giant production plants.

In the mid-1800s a machinist, like a farmer, usually made his own tools. Industrialist Arthur Briggs Farquhar recalled in a 1922 memoir that qualifying then meant mastering nearly every kind of metal work, selecting and grinding cutters, and mostly making them yourself. One theme of industrialization is how that bond broke: the factory hand no longer makes or owns his tools and stops setting the pace. When Thomas Edison began lamp production in 1880, his works were full of craftsmen who knew glass and wire and could build a lamp end to end; over time their know-how was built into specialized machinery, jobs were simplified and split. Riveting a boiler once took nine skilled men and, with a hydraulic riveter, fell to one skilled hand plus eight laborers. A young machinist named John Morrison told Congress in 1883 that the trade had subdivided since the mid-1870s, so finely in sewing-machine work that the hands there no longer counted as machinists. Numerical control continues that line, but by fusing metalworking with data it promised a far deeper shock.

So what is numerical control? A programmer reads the drawing, converts it into coordinates, and the instructions go onto paper or plastic tape feeding a controller; tool and workpiece move without the machinist guiding them by hand. Born in the 1950s, the technology served Air Force and aerospace needs first, overspecified for small commercial shops. Owners who bought NC hoping to downgrade their crews were disappointed: the gear was hard-wired and rigid, and any serious change meant opening the controller and rebuilding it. Mid-1960s surveys found shops still needed operators of equal knowledge and ability, especially during the changeover, and assigned their best hands to the costly new machines for fear of expensive downtime.

Over time buyers saw real advantages: intricate parts once impossible, high accuracy, less rework, faster runs, and shorter setups between pieces. Roughly, one NC drill replaced three conventional machines and one NC mill replaced two or three. Fewer machines meant higher utilization and better returns. Small batches turned economical, productive time rose four to five times, and robot arms for loading and unloading enabled industrial cells linked by conveyors. Pioneered in Britain and popularized in Japan, the cell combined cheap volume with product flexibility.

Change accelerated through the 1970s. By 1980 about a quarter to a third of machine tools sold in advanced economies carried NC, and competitive pressure pushed firms to stretch their budgets for the costly equipment. A major source of gains was replacing two to three times fewer veteran conventional operators with NC hands. An NC operator needed training, but nothing like an apprenticeship: a 1982 study of Swedish shops put NC training at six to twelve months against four to five years for a lathe hand. One firm reported needing only twenty-two NC operators instead of forty-four lathe men; another replaced sixty-three qualified operators with twenty-one. Running machines through odd hours paid off where skilled night-shift labor was scarce. Perception settled accordingly: in a 1980s Canadian survey nearly eighty percent of manual machinists agreed NC work took less ability, actual NC users split evenly, and managers leaned the same way as the manual men.

The microprocessor of the 1970s brought computer numerical control: instructions could be created and edited inside a computer. That partly handed design power back to the floor. Under early NC the office programmer prepared the tape while the operator only set up and watched. Manual data input panels with keyboards and screens let operators fix flaws or write their own routines without deep programming knowledge, rewarding those who paired classic know-how with the new interface. Machinist Bill Bowman, whose career spanned the change, recalled in a 1992 oral history that employers increasingly expected hands to program their own machines and that early adaptation lifted his pay; in 1989 he made about thirty thousand dollars a year at such work, roughly eighty-one thousand today.

Wage data cuts against the perception. A 1989 study of 1981 Bureau of Labor Statistics figures compared eighty thousand Class A machinists with NC operators: nine dollars seventy-two cents an hour versus nine dollars fifty-one. As a check, Class B and C hands on repetitive tasks earned about eight fifty-four and six forty-one, so NC operators recruited from those ranks effectively won a raise. The host concedes wages are a blunt gauge, shaped by place and union coverage and silent on job quality and headcount. Bowman's later years warn the same way: as the ability spread, his premium compressed until he earned twenty thousand a year, near nine dollars an hour, despite heavier programming. Still, the figures suggest NC operators were not as unskilled as they looked; mastery of one thing was traded for different abilities.

A new ability born of the shift was system supervision. Watching a machine properly means grasping its workings in the abstract; when it fails, the cause is rarely obvious and the incident must be retraced to rebuild the logic. A mid-1990s survey of Taiwanese CNC hands found eighty percent carried both machining and programming experience. A Hurco operator in an American field study told how he hears the drill laboring, reads the finish, and slows the run or enlarges the spot in the program, the same instinct once applied by easing pressure at a manual press. Yet the job moved from touch and dexterity toward screens, parameters, and messages, with layered faults demanding teamwork across specialists. Jeffrey Keefe's careful 1991 study of three official surveys over thirty years found aggregate ability down about one percent, effectively zero. But the flat average hides turbulence: setup hands, semi-skilled tenders, and laborers lost ground while tasks migrated to NC operators.

The episode closes by drawing the parallel with AI and programming. Early assistants offered simple completion, while current agentic systems build products as supervised teams of agents, the picture the host saw in a February 2026 case study by OpenAI's Ryan Lopopolo. Programming today resembles machining then: a dominant, well-paid trade. Reading and judging code will likely remain, except the fix goes to an agent as an instruction rather than into an editor; same judgment, new expression. Some roles will lose standing and fade, new fields may break out, and the NC operator of the AI era has yet to appear. The lesson the host keeps from Bowman is plain: keep learning without pause.