Which Jobs Are Actually Shrinking? What the 2026 Data Shows

Updated August 2026 — every figure below was verified or re-verified this month. Projections are the Bureau of Labor Statistics’ current 2024-34 cycle; payroll and layoff figures are the most recent available releases.

Ask which jobs AI is killing and you’ll get a hundred confident listicles. Ask which jobs are actually shrinking — in the government’s own occupational projections and in real payroll data — and the answer surprises three times. The fastest-shrinking job in America is one held by just 40,000 people, and it isn’t cashier, customer service rep, or coder. The everyday jobs losing the most positions are shrinking far more slowly than the headlines suggest. And the sharpest measured AI effect isn’t in any officially “declining” occupation at all — it’s happening inside jobs that are still growing.

Surprise One: The Fastest-Shrinking Jobs Are Ones You Forgot Existed

The Bureau of Labor Statistics’ current employment projections, released in August 2025 and covering 2024 to 2034, put word processors and typists at the top of the fastest-declining list — down 36.1% over the decade, from an occupation that only employed 40,000 people to begin with. The rest of the steepest-decline list reads like a museum of twentieth-century office work: telephone operators, switchboard operators, data entry keyers, telemarketers, file clerks.

Occupation	2024 jobs	Projected change by 2034	Percent
Word processors and typists	40,000	-14,400	-36.1%
Telephone operators	4,000	-1,100	-27.5%
Switchboard operators	36,600	-9,600	-26.3%
Data entry keyers	141,600	-36,700	-25.9%
Telemarketers	67,400	-14,900	-22.1%
Payroll and timekeeping clerks	161,100	-27,000	-16.7%
File clerks	84,300	-13,400	-15.9%

Source: BLS Employment Projections 2024-34, Table 1.5 (fastest declining occupations).

The lesson hiding in this table: steep percentage declines mostly mark the tail end of automation waves that began decades before ChatGPT. These are small occupations finishing a long fade — not evidence of a sudden AI purge.

Surprise Two: The Everyday Jobs Are Shrinking — Slowly

The list that matters to most readers is different: occupations losing the largest number of positions. Here the familiar names appear — and the percentages are smaller than you’d guess from the discourse.

Occupation	2024 jobs	Projected change by 2034	Percent	Median wage (2024)
Cashiers	3,157,200	-313,600	-9.9%	$31,190
Office clerks, general	2,646,000	-177,800	-6.7%	$43,630
Customer service representatives	2,814,000	-153,700	-5.5%	$42,830
Bookkeeping, accounting, and auditing clerks	1,613,400	-94,300	-5.8%	$49,210
Tellers	347,400	-44,900	-12.9%	$39,340
Secretaries and administrative assistants*	1,944,000	-30,800	-1.6%	$46,290
Computer user support specialists	729,500	-27,000	-3.7%	$60,340

Source: BLS Employment Projections 2024-34, Table 1.6 (largest projected job declines). *Excludes legal, medical, and executive assistants.

BLS is explicit about the mechanism: “the use of automated systems, including AI, is expected to contribute to declining employment of office and administrative support workers,” alongside e-commerce and self-service technology. But notice the scale. Customer service representatives — the job most often declared dead — are projected to decline 5.5% over ten years, and BLS still expects about 341,700 openings for the role every single year, because turnover keeps generating vacancies even as the total shrinks. A shrinking occupation and a hireable occupation can be the same occupation.

Two more numbers keep this honest. The whole US economy is still projected to add 5.2 million jobs (+3.1%) by 2034. And two of the biggest declining rows — elementary and secondary school teachers — are on the list for demographic and budget reasons, by BLS’s own attribution, not because of AI. A job appearing on a decline table tells you less than the reason it’s there.

Surprise Three: The Sharpest AI Effect Is Inside Growing Jobs

Here is the layer the projection tables cannot show — and where the “junior developers” part of this question lives. Software developers are projected to grow 15% by 2034, one of the stronger growth rates in the economy, with about 129,200 openings a year. The narrower, more routine occupation of computer programmer is projected to decline 6%, with BLS explicitly citing AI’s automation of repetitive coding tasks. Same field, opposite directions, depending on how routine the work is.

But the real evidence comes from payroll data, not projections. Stanford’s Digital Economy Lab — in the August 2026 revision of its “Canaries in the Coal Mine” study, built on ADP payroll records covering 4.6 million workers across more than 730 occupations — found that employment of 22-to-25-year-olds in the most AI-exposed occupations, software engineering and customer service among them, is running about 19% below where it would be if it had tracked less-exposed peers. That gap was roughly 15% when the study first appeared in mid-2025; a year later it has widened to 19%. The mechanism, per the study, is reduced hiring of young workers rather than firings — and the same data shows no evidence of widespread, economy-wide displacement, with employment flat or rising for experienced workers even in exposed fields.

Put plainly: the most measurable thing AI has done to the job market so far is not deleting occupations. It is quietly removing the bottom rung of occupations that are otherwise fine.

What’s Driving It Right Now

The projection tables look ten years out, but the present-tense data points the same direction. According to Challenger, Gray & Christmas — the outplacement firm whose monthly job-cut reports are the industry standard — AI was the top-cited reason for US layoff announcements in July 2026 for the fifth consecutive month: 10,970 of 33,429 announced cuts, roughly 33%. Across 2026 so far, employers have cited AI in 112,713 announced cuts, about 24% of the total. Two cautions come with that: these are announcements, not completed separations, and — as this site’s AI Layoff Tracker has documented — an “AI-cited” cut is not always an AI-caused one.

The most concrete named example sits squarely in customer service. Salesforce CEO Marc Benioff has said publicly that his company’s support workforce went from about 9,000 people to about 5,000 since the start of 2025, with AI agents now handling roughly half of support conversations and support costs down about 17% — figures that are Salesforce’s own claims, not independent audits, and Benioff notes hundreds of those workers were redeployed internally. Worth remembering alongside it: Klarna made the loudest customer-service automation claim of the decade — an AI assistant doing the work of 700 agents — and its CEO later admitted the company “went too far” and rebuilt a blended human-AI support model.

What Isn’t Shrinking

The same BLS release that projects the declines above projects healthcare support occupations growing 12.4% — the fastest of any group — and computer and mathematical occupations growing 10.1%, with data scientists the fourth-fastest-growing occupation in the economy and information security analysts rising on cyberattack demand. In one of the stranger side effects of the AI boom, the two fastest-growing individual occupations in America are wind turbine service technicians and solar photovoltaic installers, driven partly by the electricity demand of data centers.

The global picture rhymes. The World Economic Forum’s Future of Jobs Report 2025 — a survey of over 1,000 employers representing 14 million workers — expects postal clerks, bank tellers, data entry clerks, cashiers, and administrative assistants to decline fastest by 2030, while forecasting 92 million jobs displaced against 170 million created, a net gain of 78 million. That is an aggregation of employer expectations, not measured data — but it independently lands on the same clerical-and-cashier list the BLS statisticians produced, which is about as strong as cross-validation gets in labor forecasting.

One honesty note the coverage usually skips: BLS itself warns that its precise projected values carry inherent uncertainty, and that the direction and relative size of changes are the reliable signal. Treat every number above that way.

What This Means for You

The question “is my job on the shrinking list?” turns out to be less useful than three sharper ones.

First: is my occupation declining, or just changing hands — a 5% decade-long decline with 340,000 annual openings is a very different fact than a disappearing field.
Second: how much of my day is routine, repeatable, text-or-transaction work — that, not my job title, is what the Stanford data says AI is actually absorbing.
Third: am I early-career in an AI-exposed field — because the measured squeeze is at the entry rung, the practical play is to accumulate the judgment-heavy, context-heavy experience that the same data shows AI complements rather than replaces.

If you want this data tracked as it updates — BLS revises annually, Stanford is revising in near-real time, and Challenger reports monthly.