Automation and the Job Iceberg

Across sixty years of U.S. census occupation records, exactly one job title has been fully eliminated by automation: the elevator operator, last counted in 1960. Every other automation wave on record — the ATM, the spreadsheet, the word processor, autopilot — transformed or multiplied the roles it was predicted to destroy. The forward-looking question for any role is therefore not “can a machine do my tasks” but “can my role absorb the responsibility and volume that cheaper tasks create.”

The record

• The ATM doubled the tellers. As ATM counts climbed, U.S. bank teller employment climbed with them, peaking around 2007 — cheaper cash-handling let banks open more branches and moved tellers up into relationship work.
• The pilot is the counter-case to the operator. Flying has been progressively automated since the 1920s — autopilot, fly-by-wire, auto-land — and pilot demand exploded, because automation made flying cheap enough for hundreds of millions of passengers who all want a responsible human up front. The elevator operator had no equivalent: nothing to absorb, no responsibility to hold, no way to expand the role — and modern elevator dispatch outgrew what a human in the car could manage.
• 60% of current jobs didn’t exist 50 years ago. David Autor’s MIT team traced new occupation titles through 80 years of census records — solar photovoltaic electrician, cybersecurity analyst, mental health counselor — and found new work pouring in exactly where augmenting inventions land. The word processor preceded a multiplication of writers; digital editing preceded a global streaming industry.
• Tasks are not the job. Every job converts an input to an output; the button-pushing between them is the current required task set, and those tasks are frequently the part nobody likes — the lawyer wasn’t searching the case law that AI now surfaces, and Amazon re-hires 150 workers per 100 warehouse positions per year because the tasks grind people down. Lottery surveys say more than half of workers would leave their current job if money were solved; work, for most people, is an instrument for buying time and the ability to say no.
• The iceberg is the demand that doesn’t exist yet. Most things people would value are never created because the cost of creating them exceeds any visible market — a one-off 3D-printed object needs $17,000 of molds or a day of CAD skill, so “the market records nothing.” When AI collapses that cost (a few prompts and a printer run), the new output competes against non-consumption: it takes no one’s job, because no one was doing the work.

The case against the comfort

The historical argument is backward-looking, and its own presenter flags the bias. Past automations attacked narrow task bands; a general system that improves across most cognitive tasks at once has no clean precedent in the census record, so “history says augment” is an extrapolation, not a law. The pilot’s survival also leaned on physical presence, regulation, and liability — anchors many desk roles lack. And the record’s own mechanism cuts both ways: roles that cannot absorb responsibility or expand (the elevator operator’s actual failure) are exactly the roles a general automation wave finds first. The honest reading is a strong prior toward transformation over elimination, held with the tail risk named.

Boundaries

The census argument concerns job titles, not individuals — a category surviving says nothing about whether a given person’s employer retrains or replaces them, which is where the video’s operational close points: the displacement risk concentrates on workers who refuse the tools, not on the occupation. The economics also say nothing about transition pain; tellers doubled over fifty years, which is cold comfort inside any five of them.

Related Pages

• Agentic Engineering - the practice of being the augmented worker rather than the replaced one
• Understanding Bottleneck - the part of a role that automation cannot absorb: knowing what is worth building and what is true enough
• Money as Life Energy - the work-to-buy-time reading of why people hold jobs at all

Sources

• Maxinomics, “Why AI Isn’t Going to Take Your Job” — https://www.youtube.com/watch?v=zRv5kW5mAxM (transcript in raw/processed)

Open Questions

• Which current roles have the pilot’s structure (responsibility that scales with automation) versus the operator’s (a task band with no absorption capacity)?
• Autor’s data ends before general-purpose AI; what would the same census method show by 2030?