Updated August 2026 — every number below was pulled directly from its source this month.
Accountants have a job the U.S. government currently projects will grow five percent over the next decade. The same job title sits on a list of the fastest-declining careers on Earth, according to a survey of more than a thousand employers worldwide. Both of those numbers are real, current, and from credible sources. Neither one is the whole answer. That contradiction is exactly why “will AI take my job” almost never gets a straight answer anywhere else, and why this piece doesn’t try to give you one number either. Instead, it checks twelve real jobs against actual labor data, one at a time, so you can see which parts of your own work the evidence says are moving fastest, and which parts genuinely aren’t.
AI-generated illustration
What We Know (Fact)
Start with the baseline everything else in this piece gets measured against. The U.S. Bureau of Labor Statistics projects total employment across every occupation in the country will grow 3% between 2024 and 2034. That single number is the yardstick: a role growing faster than 3% is outperforming the broader labor market, and a role flat or shrinking is underperforming it, regardless of how the headlines about that role read. Two roles in this guide show the clearest, most convergent evidence of AI-driven contraction of anything checked here. Customer service representatives are projected to decline 5% over the same decade, and the Bureau of Labor Statistics’ own published materials name automated phone systems, virtual assistants, and the expanding integration of AI technologies into workflows as contributing factors — one of the few occupations where a government projection names AI directly rather than leaving the cause unstated. Secretaries and administrative assistants show essentially flat, near-zero growth, and BLS again cites AI-driven efficiency gains by name. Separately, Goldman Sachs’ 2023 breakdown of which task categories are most exposed to AI automation put office and administrative support at the top of every category measured, at 46% of tasks exposed, with legal work close behind at 44%. On the other end, the Anthropic Economic Index — which maps real Claude.ai usage against the U.S. Department of Labor’s own occupational task database — found that roughly 49% of jobs already have at least a quarter of their tasks touched by AI in some way, while only about 4% of jobs have three-quarters or more of their tasks touched. Most jobs, in other words, are partially exposed, not wholesale replaced.
AI-generated illustration
What Analysts Forecast (Forecast)
Where confirmed data stops, forecasting begins, and different institutions land in different places. The World Economic Forum surveyed more than 1,000 employers representing roughly 14 million workers across 55 economies and projects that by 2030, 170 million new jobs will be created globally while 92 million are displaced, a net gain of 78 million jobs, alongside 22% total workforce churn. The same survey projects that 39% of workers’ current skills will be transformed or made obsolete in that same five-year window. McKinsey Global Institute, in a July 2023 analysis, estimated that with generative AI, up to 29.5% of current U.S. work hours could be automated by 2030, up from 21.5% without it, and that demand for health-related and other STEM occupations could grow 17% to 30% between 2022 and 2030. Goldman Sachs, in a March 2023 report now three years old and worth treating as dated rather than current, projected that widespread AI adoption could eventually add roughly 7% to global GDP and lift U.S. productivity growth by close to 1.5 percentage points annually over a decade. PwC’s Global AI Jobs Barometer, tracked across three consecutive annual editions, documents something more immediate and less speculative: the wage premium for workers with AI skills rose from 25% in its 2024 edition to 56% in 2025 and 62% in 2026, alongside productivity growth in the most AI-exposed industries nearly quadrupling, from 7% between 2018 and 2022 to 27% between 2018 and 2024.
The Role-by-Role Risk Breakdown
Aggregate numbers hide more than they reveal. The same forces pulling one occupation up can be pulling a closely related one down, and even a single job title often bundles together tasks with completely different automation profiles. Here is what the evidence, checked this run, actually shows for twelve specific roles — not ranked against each other, since each one is assessed on its own evidence, but grouped loosely by what the data has in common.
Registered Nurses — Lower Risk
BLS projects 5% growth through 2034, faster than the economy-wide average, adding roughly 166,000 jobs on top of the current 3.39 million. Nothing in the evidence gathered this run points to near-term contraction. Physical presence, licensure requirements, and direct patient trust are the kind of task features that show up consistently, across every source checked, as the hardest to automate.
Kindergarten & Elementary School Teachers — Lower-to-Moderate Risk
BLS projects a 2% decline through 2034 — but its own materials attribute that mainly to projected shifts in student enrollment, not automation. Neither the World Economic Forum’s fastest-declining list nor Goldman Sachs’ task-exposure breakdown singles out classroom instruction the way they single out clerical and administrative work. Grading and lesson-planning assistance are documented as augmentative uses of AI in current classrooms, not headcount-reducing ones, based on the sources checked this run.
Lawyers — Moderate Risk, Sharply Bifurcated
BLS projects 4% growth, about average — but Goldman Sachs found legal work has the second-highest task-exposure score of any category measured, at 44%. Both numbers make sense together once “lawyer” is treated as two different jobs sharing one title: document review, contract drafting, and legal research are the heavily automatable side; courtroom advocacy, negotiation, and client judgment are the side every source treats as far more durable.
Paralegals & Legal Assistants — Higher Risk
BLS projects essentially flat employment through 2034, adding only about 600 positions nationally despite roughly 39,300 annual openings from turnover. This role’s actual task mix — research, document review, first-pass drafting — overlaps directly with the 44% legal-task exposure Goldman documented, more directly than a practicing attorney’s full role does.
Accountants & Auditors — Moderate Risk, Genuinely Split by Data Source
This is the clearest example in this guide of two credible, current sources disagreeing because they’re measuring different things. U.S. BLS projects 5% domestic growth through 2034. The World Economic Forum’s global employer survey separately lists accountants and auditors among occupations expected to decline. Neither figure is wrong; they come from different methodologies (a government econometric projection versus an employer-sentiment survey) and different scope (U.S. hiring versus global outlook). The pattern underneath both numbers is consistent with what’s happening in nearby roles: routine bookkeeping and reconciliation are the automatable layer, and assurance, audit judgment, and advisory work are the layer showing up as durable.
Software Developers — Strong Aggregate Growth, Sharp Entry-Level Split
BLS projects 15% growth for software developers, QA analysts, and testers through 2034 — the strongest number in this entire guide. But a Stanford Digital Economy Lab study using real ADP payroll data on millions of U.S. workers found that employees aged 22 to 25 in the most AI-exposed occupations are now employed at a rate roughly 13% to 19% below where the pre-AI trend would predict, with the gap widening in the study’s most recent update, while employment for experienced workers in those same occupations shows no comparable decline. The Anthropic Economic Index independently found that computer-and-math occupations, which cover most software engineering work, show by far the highest AI adoption of any occupational category it measured. Read together, this is a seniority split hiding inside a healthy-looking aggregate: strong news if you’re already established, a genuinely harder entry point if you’re trying to get your first developer job right now.
Customer Service Representatives — Higher Risk
Covered above under What We Know: a projected 5% decline, with BLS’s own materials naming automation and AI integration directly. Of everything checked in this guide, this is the occupation where a government source states the AI connection most explicitly.
Heavy & Tractor-Trailer Truck Drivers — Lower Risk, Near-Term
BLS projects 4% growth through 2034, and this occupation is projected to have more annual job openings, about 237,600, than any other occupation in BLS’s own comparison chart. Whether fully autonomous long-haul trucking becomes commercially and regulatorily viable at scale within the next decade is genuinely unresolved — a real open question, not a settled forecast in either direction — but it is not what today’s labor-market data shows happening.
Graphic Designers — Moderate-to-Higher Risk
BLS projects only 2% growth through 2034, slower than the economy-wide average — a notable contrast with the closely related, more technical role of web developer and digital designer, which BLS projects to grow 7%, more than three times as fast. The pattern matches what shows up across several roles in this guide: the more a job leans toward pure visual production, the weaker its number; the more it leans toward technical build work, the stronger it gets.
Financial Analysts — Lower-to-Moderate Risk
BLS projects 6% growth through 2034, faster than average. Goldman Sachs’ business-and-financial-operations category showed 35% task exposure, concentrated in the data-gathering and modeling side of the role rather than the client-facing judgment side.
Market Research Analysts & Marketing Specialists — Lower-to-Moderate Risk
BLS projects 7% growth through 2034, much faster than average — a counterintuitive result given how heavily AI tools are already used for ad copy and routine content production. The most likely explanation the evidence supports is a shift in where demand sits inside the role: toward strategy, analysis, and audience insight, and away from pure execution.
Secretaries & Administrative Assistants — Higher Risk
The most convergent evidence of any role in this guide. BLS projects essentially flat employment and names AI-driven efficiency gains directly. Goldman Sachs’ highest task-exposure category, office and administrative support at 46%, maps onto this role more directly than any other. The World Economic Forum’s employer survey separately lists administrative assistants and executive secretaries among the fastest-declining roles worldwide. Three independent sources, three different methodologies, one consistent direction.
What People Are Saying (Opinion)
Anthropic CEO Dario Amodei, in an interview published by Axios on May 28, 2025, said AI could eliminate up to half of all entry-level white-collar jobs within one to five years and could push unemployment as high as 10% to 20% in that window. Pressed on whether this framing was alarmist, he said: “We, as the producers of this technology, have a duty and an obligation to be honest about what is coming.” This is Amodei’s own stated position as the head of a company building the technology in question, not an independently measured outcome, and it should be read that way — a strongly held, publicly stated opinion, not a settled fact sitting alongside the BLS and WEF figures above.
What’s Still Speculative (Speculation)
Several of the most-repeated claims in this space are genuinely unresolved rather than quietly settled. Whether fully autonomous long-haul trucking reaches commercial and regulatory viability at scale within the next decade remains an open technical and policy question with no credible consensus timeline. Whether the World Economic Forum’s 2030 net-job-creation projection holds up is also unconfirmed, given how much AI capability itself has changed even in the two years since the underlying employer survey was fielded — a five-year forecast built on a fast-moving technology carries more uncertainty than a five-year forecast normally would. And whether the entry-level hiring contraction documented in the Stanford payroll study is a permanent restructuring of how careers begin, or a temporary adjustment period while employers work out how to use these tools responsibly, is a question the study’s own authors do not claim to answer — and neither does this piece.
What This Means for You
The honest takeaway is not “your job is safe” or “your job is doomed” — it’s that risk tracks tasks, not titles. Every source checked in this guide points the same direction on one point: the parts of any job that are repetitive, predictable, and pattern-based — routine data entry, first-pass drafting, answering the same category of question, basic scheduling and coordination — are the parts moving fastest toward automation, regardless of which industry they sit inside. The parts that involve physical presence, direct human trust, genuine judgment under ambiguity, or navigating something meaningfully new each time are the parts holding up consistently across nurses, senior lawyers, senior developers, and truck drivers alike. Almost no real job is purely one category or the other. The useful exercise isn’t checking whether your job title appears on a “safe” list somewhere — it’s an honest audit of your own actual week: which tasks are the pattern-based ones, and which are the ones that would be genuinely hard for anyone, human or AI, to do without you specifically. Build the second kind deliberately, and treat fluency with AI tools themselves as part of that second kind, not a threat to it.