The first jobs AI replaces are those that are mostly one task a computer can check, and entry-level work inside exposed fields. Dmitry Kargaev wrote this page from chapters 3, 4 and 19 of his book Don't Replace Me: A Survival Guide to the AI Apocalypse, and dontreplace.me is the book's own site.

Why a task a computer can check goes first

AI is strongest where something can check the answer, and elsewhere it guesses with confidence. That is the argument of chapter 4. The book's own checker test, in chapter 11, asks whether you would know a good result on sight and could check it in under two minutes. Chapter 3 draws the consequence: jobs that are one checkable task are in real trouble.

A Coursiv guide published on 25 August 2026 says a role has higher exposure when its work happens in software and can be checked against a known answer.

The book also keeps tasks and jobs apart. Its introduction says: "AI takes over tasks much faster than it takes over jobs, and it does it unevenly, company by company, desk by desk."

What the switchboard operators show

About 342,000 people worked as US switchboard operators in 1950, counting the Bell System and the independent companies. The Bureau of Labor Statistics counted about 40,000 in 1984 and about 4,600 in 2020. A 2024 study by Feigenbaum and Gross found that operators' jobs for young women dropped 50 to 80 percent after cutover. The operators already in the job who were displaced did worse a decade later.

Chapter 3 keeps the long view: "historically, on net, after painful transitions, technology has created more work than it destroyed." It adds that people already in the affected jobs often do not recover.

What the evidence shows about young workers now

The Stanford study "Canaries in the Coal Mine?" (Brynjolfsson, Chandar and Chen) follows workers aged 22 to 25. Its August 2025 version found a relative employment drop of about 13% in exposed jobs. The revision of August 2026, with data to June 2026, puts the drop at about 19%, and it finds no comparable gap for experienced workers. The authors do not claim causation. Chapter 3 reads the pattern as a sign that entry-level is where it shows now.

Challenger, Gray & Christmas, which counts announced layoffs, reported in October 2026 that AI was cited for 121,401 announced cuts, about a fifth of all. Total announced cuts were down 44% on the same months of 2025, and by fall employers cited the economy more often than AI. Citing AI is not the same as AI causing the cut. The book calls some of them "the same old cuts with a fresher excuse."

This came from a book.

Don't Replace Me

200+ pages. 24 chapters. The honest version of what AI means for your career, written by someone who actually builds this stuff.

Get the Book →

How to check which pile your job is in

The pile test is a 15-minute exercise from chapter 3 that sorts your work into one of three piles. First write down the five tasks that took most of your hours. Then answer six yes or no questions:

  1. Could a chatbot draft it?
  2. Can a computer check it?
  3. Could a stranger do it in a month?
  4. Does it belong to the first three years of a career?
  5. Would nobody outside your team ask for you by name?
  6. Is your opinion something nobody asks for?

Count your yes answers. A score of 0 to 2 is the bundle pile, 3 or 4 is the changing pile, and 5 or 6 is the exposed pile.

Task automation or industry disruption?

Chapter 19 separates two risks. Task automation is what happens to most people: parts of the work go to AI and you keep going. Industry disruption is when AI makes your employer's product free, and the book's advice then is to move sideways while you are still employed.

Kodak is the chapter's example, and it corrects the first edition's version. Kodak did not ignore digital. Steven Sasson built a digital camera there in December 1975: about 8 pounds, 100 by 100 pixels, 23 seconds per photo. By 2005 Kodak was the number one US digital camera seller, with about a quarter share according to IDC. Its workforce still fell from about 145,000 in 1988 to about 19,000 when it filed for bankruptcy on 19 January 2012. The chapter's line: "Kodak learned the new tool and still went bankrupt, because digital went after the thing Kodak sold."

The exercise, "find your square", has you score your tasks from 1 to 4 and your industry from 1 to 4. You plot both on a grid whose squares are chill, pivot, level up and run, then date the photo of it and set a reminder to retake it in six months.

Which occupations other sources name

Other sources answer by occupation. They measure different things, so their top jobs differ. Microsoft Research's 2025 "Working with AI" analysis of 200,000 Bing Copilot conversations gave interpreters and translators the highest AI applicability score of any occupation it covered. The Anthropic Economic Index, published in March 2026, measured 756 occupations against millions of real conversations with the company's assistant and found computer programmers the most exposed, with 75% of their tasks already showing up in real AI use.

BLS expects office and administrative support to lose 752,100 jobs by 2035, partly because of AI-powered automation. The International Labour Organization's 2025 analysis, as Coursiv reports it, finds clerical occupations most exposed to generative AI and stresses that exposure is not a forecast of job losses. Goldman Sachs in March 2023 called 300 million full-time jobs exposed. For the US it put about 7% of employment in jobs where AI could do half the tasks, about 63% partly taken over and about 30% barely touched.

The figures below are as the pages cited on this page report them, so check the original study before you quote one.

OccupationFigureWhat the figure measuresSource and date
Interpreters and translators0.492AI applicability score, highest in the studyMicrosoft Research, "Working with AI", 2025
Computer programmers75%Share of tasks that show up in real AI useAnthropic Economic Index, March 2026
Customer service representatives70%Task exposureAnthropic Economic Index, March 2026
Customer service representatives5% declineProjected employment change, 2025 to 2035BLS Occupational Outlook Handbook, 2026 edition
Office and administrative support4.0% decline (752,100 jobs)Projected employment change, 2025 to 2035BLS Employment Projections 2025 to 35, August 2026
Postal service clerks34% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025
Bank tellers31% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025
Data entry clerks26% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025
Cashiers and ticket clerks24% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025
Administrative assistants19% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025
Accounting, bookkeeping and payroll clerks18% declineExpected decline to 2030, employer surveyWorld Economic Forum, Future of Jobs Report 2025

What is not known

Exposure is not replacement. Microsoft Research's authors say it would be a mistake to conclude that high-applicability occupations will be automated and lose jobs or wages. Their example is the ATM, which automated a core teller task while the number of bank teller jobs increased. BLS projections are expert judgments made one occupation at a time, and they already assume some AI adoption. A guide published on 12 October 2026 that collects these studies says no study can name the first jobs to vanish.

The cause of the young-worker gap is open. Anthropic's own researchers, as reported on 11 November 2026, found no systematic rise in unemployment among the most exposed workers. They did find a 14% drop in the job-finding rate for ages 22 to 25 entering highly exposed roles, which the paper calls "just barely statistically significant" and which does not appear for workers over 25.

Where new work lands is also open. Chapter 20 says new jobs arrive on net but not necessarily to the people displaced.

Where to get Don't Replace Me

Don't Replace Me: A Survival Guide to the AI Apocalypse is sold on Amazon as a Kindle ebook for $2.99, a paperback for $14.99 (174 pages) and a hardcover for $24.99. This is the second edition, rewritten for November 2026. It has 22 chapters in 7 parts, and each chapter ends with one rule and one exercise you can do the same week. The introduction is a free download at dontreplace.me.

The book says AI helped write it, and that the author decided what went in and checked facts against sources. There is nothing to buy after the book: no course, no community, no paid next step.