Issue #364

What Happened to Farriers in the 20 Years Cars Replaced Horses

I lined up a 130-year-old newspaper archive against today's AI headlines—and the parallels are uncanny.

BusinessWhat Happened to Farriers in the 20 Years Cars Replaced Horses

In 1899, the New York Times Hesitated to Even Print the Word ‘Automobile’

Reader, have you ever heard AI called a toy, a world-changer, and a job-killer—all in the same day? Sometimes it’s the same person saying all three.

One English-language essay, written for an audience of developers, traces this confusion back to a remarkably old record. It’s called 「How to be a Good Farrier」, which translates to “good farrier1”—meaning someone who shoes horses. The writer went digging through old newspapers from the last time a new technology walked into everyday life and genuinely frightened people: the period when the automobile was pushing the horse aside. The essay’s starting point is this: what people said back then and what people say about AI now unfold in almost exactly the same sequence.

Let me set the scene. In a January 3, 1899 editorial, the New York Times wrote that this new mode of transport was uniformly hideous and didn’t even have a proper name yet. The word “automobile,” coined by the French, was half Greek and half Latin—practically vulgar, the paper said—adding that it was being printed “with some hesitation.” A newspaper apologizing to its readers for putting a single word on the page.

At this point, the essay’s author asks you to picture today’s people grumbling about the word “agentic.” I think this one scene compresses the entire essay. People meeting a new technology for the first time got uncomfortable with its name before they got uncomfortable with the technology itself.

When Horses Ran the City, the Car Was a Rich Man’s Toy

According to records compiled in the essay, roughly 130,000 horses worked in Manhattan alone in 1900. New York City hauled away nearly 15,000 dead horses in 1880 — about 41 a day. The 1900 US census counted 226,477 blacksmiths, 361,770 teamsters, and 64,929 stable hands. Horses needed new shoes every few weeks, so even a mediocre farrier never ran short of work.

Early automobiles, by contrast, were nothing like everyday tools. A Columbia automobile cost $3,500 in 1901, when the average worker’s annual salary was $454 — meaning you’d have to save nearly eight years of wages, spending nothing, just to afford one. In 1903, the best-selling budget car, the Oldsmobile, moved just over 4,000 units; that same year, more than 900,000 people bought a horse-drawn carriage.

So it’s no surprise that the sharpest minds of the era wrote off the automobile as a fad. In 1899, the Literary Digest declared that even if car prices eventually fell, they would “of course” never become as common as bicycles. The essay cites an anecdote from a 1922 biography of Ford: in 1903, a Detroit banker warned Henry Ford’s lawyer against investing, telling him the horse was here to stay and the automobile was just a passing craze.

The objections always arrived in the same order

People didn’t stop at being suspicious of cars—they tried to stop them outright. In 1894, Vermont passed a law requiring that any steam car on a public road send an adult ahead on foot, an eighth of a mile in advance, to warn people. The law lasted 6 years. In 1898, Mackinac Island, Michigan, banned cars after a petition from carriage drivers—a ban that remains in effect to this day. Nantucket blocked cars until 1918, and in 1907 the New York Times reported that the real reason, according to car owners themselves, was to protect the stable-rental business that employed much of the island’s population.

At a hearing in November 1899 over whether to allow cars into Central Park, one stable owner testified that the park should be preserved for family carriage rides, and that cars, above all, were too expensive for ordinary people. A lawyer for the carriage drivers’ union shot back that he’d heard of horses breaking down, but never of one exploding. Dangerous. A rich man’s toy. Liable to blow up. Every person making these claims earned their living from horses.

In July 1913, the New York Times ran a piece headlined “Passing of Horse Only Myth of Auto Trade.” A horse-trade magazine editor was quoted calling the “horseless age” as distant as the millennium. According to the essay, the number of horses on American farms peaked within 5 years of that article—and never recovered.

Run AI through the same sequence and you get this. The “toy” line shows up in Ted Chiang’s February 2023 New Yorker piece, where he describes ChatGPT as a blurry JPEG of the entire web. The bans came in the form of New York City’s public schools blocking access just 5 weeks after ChatGPT’s launch, Italy’s month-long nationwide ban, and the March 2023 open letter calling for at least a 6-month pause on training any model more powerful than GPT-4. On March 25 this year, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez introduced a bill to halt construction of new AI data centers until national-level safeguards are in place. And the job-loss panic surfaced in a 2023 Senate hearing, where Senator Richard Blumenthal named the loss of millions of jobs as his biggest nightmare.

The essay’s author isn’t saying these people were all wrong. It was true, in 1902, when the New York Times criticized wealthy speeders for hitting children—cars still kill people today. The author simply draws a line between “there’s a problem” and “it can be stopped”—two entirely different claims.

When the price dropped, the curve bent

What turned the automobile from a toy into a tool of daily life wasn’t editorials or hearings — it was price. Ford launched the Model T on October 1, 1908, and the base price and production volume for the cheapest body style, the Runabout, moved like this.

YearModel T Runabout PriceUnits Produced
1909$82510,666
1912$59068,733
1914$440202,667
1916$345501,462
1925$2601,911,705

With the moving assembly line introduced in 1913, the time to build a single car dropped from 12.5 hours to 93 minutes. According to the US Federal Highway Administration, registered vehicles grew from 8,000 in 1900 to 468,500 in 1910, to 9.23 million in 1920, and to 26.75 million in 1930. The share of households owning a car went from 10% in 1915 to 54% in 1926. Not a single editorial, island-wide ban, or state legislature petition bent that curve even slightly.

The numbers in AI today are stranger. Here’s Stack Overflow’s annual developer survey, year by year.

YearUsing or Planning to Use AI ToolsFavorableDon’t Trust Accuracy
202370%77%27%
202476%72%31%
202584%60%46%

Usage keeps climbing year over year, while liking and trusting it keep falling. In the 2025 survey, about half of professional developers used AI tools every day. GitHub’s annual report, Octoverse 2026, released on October 28, found that over 40% of pull requests2 this September started from a draft written by an agent — up from 17% in March. A Gallup survey found that as of May this year, 52% of US workers had used AI at work at least a few times over the past year.

In 1902, The New York Times wrote that the automobile had first been an object of fear and had quickly become an object of widespread loathing. U.S. motor vehicle registrations rose from 78,800 in 1905 to about 2.5 million in 1915. Trust falling while usage rose — we’ve seen this movie before.

So What Happened to the Farrier

This is the question the essay spends the most effort answering. The writer’s first answer is fairly blunt: the odds were that you’d lose your job. Here are the numbers the essay pulls from the census occupation tables.

Occupation191019201930
Blacksmiths (including farriers)232,988195,255124,373
Teamsters (horse-drawn cart drivers)443,845420,189111,224
Stable hands63,38818,9766,654
Drivers & truck drivers45,785285,045972,418
Garage operators5,27942,15169,965

Over twenty years, nine out of ten stable hands disappeared, three out of four teamsters disappeared, and half of all blacksmiths disappeared. But you have to read the bottom two rows together. While teamsters declined by 330,000, drivers increased by over 920,000. Since the census counts jobs, not people, you can’t trace the path of any individual who moved from a carriage seat to a truck’s driver’s seat. But the fact that the Census Bureau at the time complained about how hard it was to cleanly separate “teamster” from “driver” is itself evidence of how closely the two jobs were actually linked.

Some blacksmiths made the jump too. In 1934, the Reading Times, a local Pennsylvania newspaper, wrote that the blacksmiths of the day were making a living by straightening car axles, fixing springs, and welding chassis frames. The Fisher brothers, who came from a family of blacksmiths building carriage bodies, founded Fisher Body in 1908 and went on to build car bodies for Cadillac and Buick.

The same Reading Times article notes that the city had only about 150 horses left and just three people shoeing them. The article doesn’t say who those three were. The writer guesses that when there’s only enough work for three people, the job goes to whoever is best at it—or best connected. Probably not the guy who injured a horse’s hoof twice last year and kept showing up late.

This job didn’t disappear. There’s no official statistic, but the American Farriers Journal, a trade publication for American farriers, estimates that roughly 27,500 farriers—full-time and part-time combined—are caring for about 7 million horses in the US. Given that farms alone had over 21 million horses in the late 1910s, the market has clearly shrunk a great deal. And yet, according to the journal’s latest survey, the average annual gross income for a full-time farrier, before expenses, is $135,736. No license required. You just have to be good enough that people with expensive horses keep calling you back.

What the Data Says Now — and What It Doesn’t

The essay argues this is already happening. There’s a study called “Canary in the Coal Mine,” run by Stanford’s Digital Economy Lab, which tracks data from ADP, a major US payroll processing company. In its August 12 revision, the study found that employment for 22–25 year-olds in AI-exposed occupations3 is 19% lower than it would be if it had followed the same-age trend in low-exposure occupations. No comparable gap shows up among experienced workers, and the difference shows up as reduced hiring rather than layoffs. The researchers are careful to note that they don’t see large-scale job displacement across the economy as a whole, and that these numbers are a descriptive pattern, not a causal estimate.

Job postings point in a similar direction. According to Indeed Hiring Lab, 69.3% of US software developer postings in Q1 of this year were for senior roles, while only 4.5% were entry-level. The same Hiring Lab also found that software developer postings have risen nearly 15% since late February 2025 — and that 71% of that increase was in senior positions. In other words, jobs aren’t simply disappearing; they’re rebounding, but skewed toward experienced hires.

What I found most convincing in the essay was where the writer draws the line on their own claim. What this data measures is age and tenure, not skill. Payroll records have no column for “an average employee.” Software job postings had already been declining since 2022 — before ChatGPT even existed — and the Stanford researchers themselves say AI isn’t the only cause. Indeed points to the spread of remote work, high interest rates, and the correction of pandemic-era overhiring as candidate explanations for weak entry-level demand. So the writer is explicit: the claim that “good performers survive” isn’t a research finding — it’s their own reading of history.

good farrier car ai jobsBuilding on that reading, here’s the advice the writer offers. Know the work itself deeply, not just the prompts. Don’t reject the new tools — use them better than the person next to you. Take responsibility for the outcome, not just the output. Move early toward the side that’s doing the replacing. Don’t wait for someone to forbid it for you.

The evidence behind the first piece of advice is interesting. In a 2025 experiment by the nonprofit research organization METR, 16 experienced developers actually worked 19% slower when using AI tools — yet they believed they’d worked 20% faster. The writer cautions against reading this as a current benchmark, since it measured 16 people using early-2025 tools, but insists the gap between perceived and actual performance is worth holding onto regardless. The point: you need to be able to do the work yourself in order to judge whether the machine actually helped.

Oswarld’s Lens

I call myself an AI optimist. So I don’t have much reason to disagree with this essay’s conclusion that AI “can’t be stopped.” When I design enterprise deployments or AI-adoption consulting projects, I work from the premise that models always get better. When I’m setting up closed networks for public institutions, I plan the project around roughly what level of open-weight model will exist 6~12 months out. The advice to not reject the tool but to use it better than the person next to you points in the same direction as the work I do in the field.

Still, I see one gap this essay leaves open. The three farriers left in Reading in 1934 were already skilled people. But saying the data is shrinking now isn’t about the skilled — it’s about the entry point for newcomers. If entry-level software postings make up 4.5% of listings on Indeed, where will the “three best” of ten years from now have learned their craft? “Be a good farrier” is sound advice for an individual, but who preserves the apprenticeships where good farriers are made is a question the organization has to answer.

In the August issue on the Luddites, I wrote that the Luddites weren’t fighting machines — they were fighting a system that billed the cost of change only to them. I don’t think that piece and this essay contradict each other. That technology can’t be stopped, and that we must reckon with who pays its cost, can both be true at once. Coachmen could become truck drivers partly because driving was something you could learn in a few weeks. Whether the current transition can be crossed by a bridge that short, nobody knows yet.

Closing

The essay’s last line is short: the horse never came back, but the good farrier kept his trade. I want to add just one line to that. Every good farrier was once someone’s apprentice, too.


💬 In the work you do now, what was the first task AI took away? Once that task disappeared, did what remained become harder, or easier? Let me know in the comments.

📨 If you have a junior colleague or peer wrestling with career anxiety because of AI, pass this piece along. The 1910 census form says more than it seems to.

Your take shapes the next issue

What resonated most in this issue, or where has your experience been different?

Any registered reader can comment for free.

References & Further Reading

Data

No Widespread Displacement, but the AI Employment Gap for Young Workers Has Widened to 19%digitaleconomy.stanford.edu

Related past issues worth reading


Illustrated portrait of Kwangseob Ahn (Oswarld)

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

  1. Farrier: A craftsperson who trims horses’ hooves and fits iron horseshoes. In earlier eras, blacksmiths often did this work too, which is why census records sometimes grouped farriers under the blacksmith category. ↩

  2. Pull request: A unit by which a developer asks for a proposed code change to be merged into a project. Once another person reviews and approves it, it gets incorporated into the main codebase. ↩

  3. AI-exposed occupation: A job in which a large share of the work can be performed or substantially assisted by generative AI. Software development, customer support, and accounting are common examples. ↩