The Job-Killing Math Nobody Taught You: Ten Collapses, One Equation, and Which Side of History You're Standing On
Every job is a bet on two prices: what the tools cost, and what the judgment to use them costs. Hand-loom weavers, switchboard operators, pin boys, and typesetters all lost both bets at once — and vanished. Bank tellers lost only one — and split into a smaller, better-paid role instead. Ten documented cases spanning 1785 to 2020 show exactly which side of that line every profession in 2026 is standing on, and why AI is not a twist on the pattern — it's the first technology to attack all of them simultaneously.
The Thesis
Strip away the job title, the credential, and the industry, and one structure sits underneath every paid occupation that has ever existed: someone pays for an outcome, and producing it requires exactly two things — a tool, and the judgment to operate it correctly. Marketing, credentials, and union cards decide who gets to supply those two things. They do not create the value. They allocate it.
From that alone, the entire history of automation follows a predictable rule: when only one input's price falls, the money flows to whoever still controls the other, scarcer input. When both prices fall together, there is no bottleneck left for the money to flow toward — and the wage for the whole profession collapses at once. This is not a new theory. It is a description of ten already-completed cases, dated and named, from 1785 to 2020. Every one of them predicted its own outcome in advance, if you knew which axis to watch.
The Only Equation That Matters
Label the two inputs technology price and intelligence price, each either expensive or cheap, and every profession that has ever existed sits in one of exactly four boxes at any given moment. The question that decides a career is not "will AI affect my job?" — it's "which of these four boxes am I in, and which direction is the wall moving?"
| Intelligence Expensive | Intelligence Cheap | |
|---|---|---|
| Technology Expensive | Q1 — The Frontier: fusion engineering, frontier AI research, novel oncology drug discovery. Billion-dollar infrastructure paired with judgment that has no textbook yet. | Q3 — Capital-Gated: airline pilots, semiconductor fab technicians, nuclear operators. Liability and licensing keep humans employed after judgment got automated. |
| Technology Cheap | Q2 — The Old Boom Zone, Now Splitting: electricians, plumbers, dentists (embodied, untouched) and senior counsel, staff engineers (informational, safe only at the top). | Q4 — Where 2026 Layoffs Concentrate: junior coding, first-draft copywriting, document review, scripted support. 265,000-and-climbing AI-attributed cuts in 2026 alone. |
The Case That Won't Sit Still
Diagnostic radiology belongs in this map honestly, not comfortably. The capital asset is genuinely expensive — an MRI or CT scanner runs into the millions, which argues for Q3. But the actual task, pattern-matching an image against known presentations, is close to the single most AI-tractable task in medicine, which argues for Q4. Radiology isn't confidently placed in either box. That ambiguity is the point: it's a profession being fought over between two quadrants in real time, not a settled case.
Exhibit A · Ten Dated Collapses, 1785 to 2020
The Pattern Is Not a Theory. It Already Happened Ten Times.
1785-1816
Hand-loom weaving — the origin of 'Luddite'
Cartwright's power loom (1785) and Hargreaves' spinning jenny (1764) cut both the machinery cost and the years of apprenticeship needed to operate it — the same workers, the same collapse, at the same time. The Luddite riots of 1811-16 were a correctly-reasoned response to a double collapse, not an irrational reflex. The word survived. The trade did not.
1852-1970s
Elevator operators — the strike that backfired
Otis proved his safety brake in 1854 by riding a platform and cutting the rope with an axe in front of a crowd. For 90 years a human operator stayed employed purely on public trust, not judgment. The 1945 NYC strike (15,000 workers, 250,000+ allied union members) forced building owners to prove automatic elevators worked — accelerating the exact automation the strike was trying to prevent.
1878-1984
Telephone switchboard operators
Strowger's automatic exchange opened in 1892; Bell moved slowly (only a third automatic by 1930). Peak employment: 342,000 in 1950, overwhelmingly women. By 1984: 40,000. Bryant Pond, Maine ran the last hand-cranked exchange in the US until 1983.
1936-1958
Pin boys and the Automatic Pinspotter
AMF's mechanical pinsetter, unveiled 1946, gained commercial acceptance by 1952. By 1958, 40,000 machines were leased nationwide. 'Pin boy' went from a common teenage job to a historical curiosity in under a generation.
1884-1987
Hot-metal typesetting
The Linotype (1884) mechanized the tool but not the skill — operators stayed well-paid and unionized for a century. Desktop publishing (Mac 1984, PostScript 1985, PageMaker 1985) collapsed both axes at once. ITU membership halved 1984-87; the union merged into CWA in 1987 after 134 years independent.
1880s-1960s
Human computers — the one case that ends in survival
Women at Harvard Observatory and NASA Langley (later 'Hidden Figures') performed calculation by hand until ENIAC (1945) and the IBM 704 (1954). Dorothy Vaughan taught her NASA team FORTRAN as the machines arrived — moving from the profession being eliminated into the profession being created by the same technology.
1990s-2012
Print encyclopedias and door-to-door sales
Encarta (1993) cheapened the technology of information storage. Wikipedia (2001) then cheapened the intelligence side — curation became free volunteer labor. Britannica printed its final edition in 2012, 244 years after its first.
1995-2007
Travel agents — a single-axis case
Expedia, Travelocity, and Priceline cheapened one narrow form of intelligence: comparing prices across airlines. ASTA membership fell roughly 50% within a decade; industry employment fell 60% from its 2000 peak by 2007 — a partial thinning, not a full collapse, consistent with only one axis moving.
1980s-2020
NYSE floor traders — the most recent case
Electronic trading displaced the technology axis through the 80s and 90s. The floor persisted in ceremonial form for decades — until COVID forced fully electronic execution on March 23, 2020, formally acknowledging what had already been true for years.
Bank tellers are the cleanest example of a profession splitting instead of vanishing
Confirmed — The Model CaseSources: ATM installation history, Barclays 1967; US banking employment records
Exhibit B · Why This Isn't Case Eleven
Every Prior Collapse Hit One Trade. This One Doesn't.
Every case above shares a structural limit 2026 does not have: each one was confined to a single tool-plus-skill combination inside one trade. Automatic switching only ever threatened the judgment of routing a phone call. Desktop publishing only ever threatened the judgment of setting type. What's being cheapened now is the general capacity to read, reason in natural language, draft, and produce a plausible answer — a capability sitting underneath law, accounting, programming, journalism, financial analysis, and customer support simultaneously.
54,836
AI-cited job cuts, all of 2025
101,743
AI-cited cuts through June 2026 alone
March 2026
AI becomes #1 cited layoff reason overall
~265,000
Full-year 2026 AI-attributed cuts (tracking)
No prior technology in the historical record touched this many professions' scarce input in the same window
ConfirmedSources: Tech-Insider AI layoff tracker; TechRadar, March 2026 layoff-reason ranking
Exhibit C · The Split Nobody Saw Coming
The Lawyer and the Plumber Were Never in the Same Boat
Every professional living in cheap-technology, expensive-intelligence territory for the last half-century assumed they were protected by the same logic. They weren't — because "intelligence" was never one thing.
Embodied Intelligence
Electricians, plumbers, HVAC technicians, dentists, physical therapists
A plumber's judgment depends on diagnosing a leak by sound and feel, then physically executing a fix in a crawlspace no camera has mapped. Large language models have made almost no progress on this kind of intelligence, because it depends on robotics and manual dexterity — a separate technology stack that hasn't had software's cost collapse. A general-purpose robot with ordinary human manual competence is still not economically deployed at scale as of 2026. No case in the historical record shows this bracket undergoing a double collapse.
Informational Intelligence
Corporate lawyers, staff engineers, portfolio strategists
Read the facts, apply precedent, produce an argument — entirely inside a document. This is precisely the kind of intelligence large language models have gotten extraordinarily capable at. Informational Q2 is safe, but only at the tier reached after the climb — and the junior tasks that used to build someone into that tier are exactly the Q4 tasks now being automated.
The Ladder Is Being Removed From Underneath the Top Floor
The document-review, first-draft-memo, junior-analyst tasks that used to train a person into senior counsel or staff engineer are exactly the tasks now being automated. This is the direct cause — not a side effect — of 2026 venture capital diligence explicitly screening for founders who already combine deep technical depth with domain expertise, rather than backing raw, fast generalist talent the way it did for the software boom's first fifty years. The top of the ladder stays valuable. The rungs are being sawn off.
What This Implies, By Name
Embodied Q2 — electricians, plumbers, dentists, aircraft mechanics
Strongest Historical SupportQ3 — pilots, semiconductor fab technicians, nuclear operators
Real, But on Borrowed TimeInformational Q2 — senior counsel, staff engineers, principal physicians
Safe Only After the ClimbQ4 — junior coding, first-draft copywriting, document review, scripted support
No Recorded Recovery From InsideThe One Documented Way Out
"Dorothy Vaughan, who led NASA's West Area Computing unit, taught herself and her team FORTRAN in the early 1960s as the electronic machines arrived — moving the group from the profession being eliminated into the profession being created by the same technology."
— NASA West Area Computing history; Hidden Figures
It is the cleanest historical example available of a group correctly reading which quadrant a falling technology price was about to create, and moving into it before the old one closed. Every other profession that landed in the low-technology, low-intelligence quadrant and stayed there — hand-loom weavers, switchboard operators, pin boys, hot-metal typesetters — has no recorded recovery. The ones that survived moved first.
Case Status: Open
Confirmed by the Historical Record
- Ten documented double- or single-axis collapses, 1785-2020, with named technologies and dates
- Bank tellers split rather than vanished when only one axis fell
- No profession has recovered a lost input from inside the low-tech, low-intelligence quadrant except by moving to a new quadrant entirely
- AI-attributed layoffs nearly doubled the prior full year by June 2026 alone
- March 2026 was the first month AI became the single most-cited layoff reason across the whole labor market
Genuinely Unresolved
- Diagnostic radiology — capital-expensive and intelligence-cheap arguments both apply
- Whether Q3 professions (pilots, fab technicians) face a later technology-axis collapse the way elevator operators did
- The exact size and timing of the "265,000" full-year 2026 projection — a tracking estimate, not a closed count
Not Yet Tested
- Whether many professions losing their scarce input in the same 18-month window behaves differently than the historical one-trade-at-a-time pattern
The Punchline
Two Prices, Not One Story
Every one of the ten collapses above followed the same equation: watch which price is falling, tool or judgment, and watch whether the other one is falling with it. When both fall together, there is no bottleneck left for the money to flow toward, and the wage collapses — not because the work was low-status, but because the specific combination it once required no longer has to be bought from anyone in particular.
What makes 2026 different isn't the equation. It's that the equation is now running underneath dozens of professions at once, for the first time in the historical record — and the only documented exit, Dorothy Vaughan's, was never about resisting the collapse. It was about reading the map correctly and moving before the door closed.
References
[1] Tech-Insider — AI-attributed layoff tracking through 2026 (54,836 cuts in 2025; 101,743 through June 2026).
[2] TechRadar — March 2026: AI becomes the #1 cited layoff reason across the labor market; Gartner ROI findings.
[3] Richmond Fed, Econ Focus Q4 2019 — "Goodbye, Operator": telephone switchboard operator employment history.
[4] HISTORY.com — The Rise and Fall of Telephone Operators.
[5] Conversable Economist — Telephone Switchboard Operators: Rise and Fall.
[6] National Inventors Hall of Fame — Almon B. Strowger; Elisha Graves Otis.
[7] Ancestry.com / NPR / LinkedIn (Henry Greenidge) — the 1945 New York elevator operators' strike.
[8] Smithsonian Lemelson Center — "Set 'Em Up, Knock 'Em Down": AMF's automatic pinsetter.
[9] UPI Archives, November 1986/1987 — International Typographical Union merger with CWA.
[10] The Conversation — the rise and fall of the typesetters' union.
[11] Stripe Economics (Ernie Tedeschi) / Edward Conard — the decline of travel agents.
[12] Wikipedia — Floor trader; NYSE floor trading suspension, March 2020.
[13] Startupfeed — 2026 seed round sizing for AI-native startups.
[14] Edmund Cartwright power loom patent (1785); Richard Hargreaves spinning jenny (1764); Luddite riots 1811-16.
[15] Elisha Otis elevator safety brake demonstration, New York Crystal Palace Exhibition, May 1854.
[16] Almon Strowger automatic switching patent (1891); first Strowger exchange, La Porte, Indiana, Nov 3, 1892.
[17] Gottfried Schmidt / AMF Automatic Pinspotter, unveiled American Bowling Congress Tournament, Buffalo, 1946.
[18] Ottmar Mergenthaler Linotype machine (1884); Apple Macintosh (1984), Adobe PostScript (1985), Aldus PageMaker (1985).
[19] Dorothy Vaughan and NASA West Area Computing unit; ENIAC (1945); IBM 704 (1954).
[20] Microsoft Encarta (1993); Wikipedia (2001); Encyclopaedia Britannica final print edition, 2012.
[21] Expedia (1996), Travelocity (1996), Priceline (1998); American Society of Travel Agents membership data.
[22] Barclays first ATM installation, London, 1967; US bank teller employment history.
This report synthesizes documented labor history and publicly reported 2025-2026 layoff data available as of September 2026. The four-quadrant framework is this report's own analytical model applied to historical cases; it is offered as a pattern with strong historical support, not as a certainty about any individual profession's future. Where a claim is labeled Confirmed, it is independently verifiable through the cited sources. Where labeled Unverified or Contested, it represents this report's interpretation of an open or ambiguous case.
Start a Critical Discussion
These questions don't have consensus answers. Share one to LinkedIn or X and see what your network actually thinks.
"Hand-loom weavers, switchboard operators, pin boys, and typesetters all lost the price of their tools AND the price of their judgment at the same time — and none of them recovered from inside that quadrant. Is 'reskill' realistic advice, or does history say the only real exit is Dorothy Vaughan's: move before the door closes?"
"Bank tellers didn't disappear when ATMs arrived — they split into a smaller, better-paid 'personal banker' role, because only the low-judgment half of the job went cheap. Is that the honest model for what happens to junior lawyers and junior engineers right now, or is AI different because it's cheapening the judgment too?"
"Every automation wave in history — the loom, the switchboard, the Linotype — hit exactly one trade at a time. AI is cheapening the same underlying skill (reading, reasoning, drafting) across law, coding, journalism, and finance simultaneously. Has that ever happened before, or is 2026 actually unprecedented in the historical record?"
"A plumber's judgment is embodied — diagnose by feel, fix in a crawlspace no camera has mapped. A corporate lawyer's judgment is informational — read, apply precedent, argue, all inside a document. AI has made almost no progress on the first and extraordinary progress on the second. Does that mean the trades just won the next fifty years by accident?"
Share this analysis
If this changed how you think about something, share it. The AI workforce conversation needs more data and less hype.