Which Jobs Will AI Replace by 2030? A Sourced Occupation Table

How the headline 300 million number actually works

The widely circulated figure that generative AI could 'expose' 300 million full-time-equivalent jobs comes from a single Goldman Sachs Global Investment Research note dated March 26 2023, written by Joseph Briggs and Devesh Kodnani. The note maps O*NET task data to a measure of how much of each task generative AI could automate, then scales the result by employment in each occupation across the United States, Europe and the rest of the world. The 300 million number is a global aggregate of full-time-equivalent task hours, not 300 million people who will lose their jobs.

Goldman's same note estimates that roughly two-thirds of US occupations are exposed to some degree of automation by generative AI, and that AI could substitute for up to one-quarter of current work in the United States and Europe. The authors specifically argue that, historically, automation has displaced workers in some tasks while creating new occupations in others — the net employment effect over a decade is therefore much smaller than the gross exposure number suggests. When you see '300 million jobs' in a headline, the careful read is 'task content equivalent to 300 million full-time jobs that an LLM could plausibly do part of.'

If you want the same task-level methodology applied to your own occupation, the Eloundou et al. paper publishes per-occupation exposure scores in its appendix, and the calculator at AI productivity multiplier by role lets you enter your weekly task mix to see what proportion is in the high-exposure bucket.

Where the research agrees: office, legal, STEM and creative roles lead the exposure list

Goldman Sachs, McKinsey and Eloundou et al. use different methods but produce surprisingly similar occupation rankings. Goldman's industry-level breakdown puts office and administrative support at 46% of task content automatable, legal at 44%, architecture and engineering at 37%, life, physical and social sciences at 36%, and business and financial operations at 35%. McKinsey's July 2023 report finds that by 2030, the midpoint of its scenario range has 29.5% of current work hours in the United States automated when generative AI is layered on top of previously-modeled automation technologies — up from 21.5% in its pre-generative-AI 2030 baseline.

Eloundou et al. take the most granular approach: they label each of the roughly 19,000 tasks in O*NET on whether GPT-4-class models could reduce the time to complete that task by at least 50% while preserving quality. They then aggregate to occupations. Their headline result is that around 80% of US workers could see at least 10% of their tasks affected, and about 19% of workers could see at least 50% of their tasks affected. The list of occupations where 100% of tasks are exposed includes mathematicians, tax preparers, financial quantitative analysts, writers and authors, web and digital interface designers, accountants and auditors, news analysts, public relations specialists, and proofreaders.

What none of the three sources claim is that these occupations will disappear by 2030. McKinsey's central scenario, for example, projects 11.8 million occupational transitions in the United States by 2030 — workers moving roles — rather than 11.8 million net job losses. The distinction is crucial when you read the same data through a personal-career lens rather than a macroeconomic one.

Where the research disagrees: blue-collar and frontline work

The biggest methodological split between sources concerns physical and frontline work. Goldman Sachs models exposure across all major occupational categories and finds that physical-labor-heavy categories — building and grounds cleaning, construction and extraction, installation, maintenance and repair, production, transportation — have automation potential between 1% and 28%, with the bottom of that range applying to roles dominated by manipulation and mobility tasks that current language models cannot perform.

Pew Research's 2023 analysis 'Which U.S. Workers Are More Exposed to AI' takes a different framing. Pew distinguishes between jobs where AI is 'more likely to help than hurt' (typically higher-paid, more-educated workers whose tasks AI can assist) and jobs where AI exposure is high but the work cannot be easily automated end-to-end. Their analysis finds that women, Asian, college-educated and higher-paid workers are most exposed to AI in this 'help' sense, which is almost the inverse of the pre-LLM automation literature that focused on routine manual tasks.

The OECD's AI Working Paper series (Lane, Williams, Broecke and others, 2023–2024) reaches a similar conclusion for OECD countries: high-skill occupations have the greatest AI exposure, but exposure correlates with productivity gains more often than with layoffs. For a deeper read on which roles the same literature flags as resistant, see jobs most resistant to AI in 2026.

What McKinsey's 2030 scenario actually projects for the United States

McKinsey Global Institute's July 2023 report is the most concrete 2030 projection in the mainstream literature. Its midpoint scenario, which assumes generative AI is adopted at a moderate pace, projects that 30% of work hours in the United States could be automated by 2030, up from a pre-generative-AI projection of 21.5%. The largest absolute gains come from four occupation families: office support, customer service and sales, food services, and production work — together accounting for roughly 84 million of the projected 11.8 million net occupational transitions and over half of the additional automation potential generative AI adds.

Within knowledge work, McKinsey calls out three use cases as having the largest impact: customer operations (chatbots, agent assist), marketing and sales (content generation, personalization), and software engineering (code generation and review). They estimate generative AI could add the equivalent of $2.6 trillion to $4.4 trillion in annual economic value across all use cases globally — a productivity number rather than a layoff number. For an industry-by-industry breakdown of when each of these use cases is likely to scale, see AI replacement timeline by industry in 2026.

McKinsey is explicit that 30% of work hours automated does not mean 30% of jobs eliminated. Historically, when automation removes tasks from an occupation, employers typically redirect workers to higher-value tasks within the same role rather than cutting headcount one-for-one. The 11.8 million occupational transitions figure is the more useful planning number for individuals and HR teams.

Are software engineers safe from AI by 2030?

Software developers (BLS SOC 15-1252, roughly 1.7 million US workers per BLS OEWS May 2023) sit in an unusual position. McKinsey identifies software engineering as one of the three highest-impact generative-AI use cases. Eloundou et al. find that programming-related tasks are among the most automatable by GPT-4-class models. Yet the BLS Employment Projections program, in its 2023–2033 release, still projects software developer employment to grow 17% between 2023 and 2033 — much faster than the all-occupations average of about 3%.

The reconciliation in the literature is that AI coding tools change the task mix and raise per-developer output, but demand for software keeps growing fast enough that headcount keeps rising. The World Economic Forum's Future of Jobs Report 2025 lists 'AI and machine learning specialists' as one of the fastest-growing job categories worldwide through 2030, alongside data analysts, big-data specialists and cybersecurity roles. The same WEF report projects a net positive employment effect from AI and information processing technologies overall, even after accounting for displacement.

The honest answer for an individual developer is therefore: the role is heavily exposed in the task-level sense, headcount in aggregate is still projected to grow, and the gap is closed by per-developer productivity rising sharply. For a sense of how much time AI coding tools actually save on real engineering tasks, see how much time AI saves by task in 2026.

Which jobs are most at risk: the four-source consensus

Cross-referencing the four primary sources — Goldman Sachs, McKinsey, Eloundou et al. and Pew Research — the occupations that appear on every short list are: tax preparers, bookkeeping and accounting clerks, paralegals, telemarketers, customer service representatives, market research analysts, technical and copy writers, proofreaders, translators, and word processors and typists. These are roles where the core deliverable is text or numerical analysis that an LLM can draft to near-finished quality, and where the volume of routine work is high enough that even partial automation removes a meaningful share of hours.

Brookings' 2019 'Automation and Artificial Intelligence' index, updated in subsequent Brookings notes through 2024, reaches similar conclusions for the white-collar slice of the labor market, though Brookings' broader index includes routine manual roles that current language models do not directly threaten. The OECD AI WPS series, drawing on a different survey methodology, places roughly 27% of OECD jobs in the highest-automation-risk category, with finance, manufacturing and ICT having the highest concentration of high-risk roles.

On the question of which specific tools matter most for each profession, the companion guide best AI tools by profession in 2026 maps each high-exposure role to the LLM stack the underlying research actually tested against.

What the World Economic Forum's 2025 Future of Jobs Report adds

The World Economic Forum's Future of Jobs Report 2025, published in January 2025, is the most recent large employer survey on AI's labor-market impact. The report surveys more than 1,000 employers covering roughly 14 million workers across 22 industry clusters. Its central finding is that 41% of employers worldwide plan to reduce their workforce in roles where AI can automate tasks, while 77% plan to upskill and reskill their existing workforce. The net employment impact of structural labor-market transformation is estimated at +7% of total employment — a positive number — but with significant churn underneath.

WEF's fastest-growing roles through 2030 include big data specialists, fintech engineers, AI and machine learning specialists, software and applications developers, and security management specialists. The fastest-declining roles include postal service clerks, bank tellers, data entry clerks, cashiers, and administrative assistants — substantially the same office-support cluster that Goldman, McKinsey and Eloundou et al. identify from the exposure side. The consistency across methodologies is unusually high for labor-market forecasting.

WEF also publishes an updated estimate of the skill stability gap: roughly 39% of workers' core skills are expected to change between 2025 and 2030, with analytical thinking, AI and big data, and technological literacy ranking as the three fastest-growing skills employers say they need.

How to read this table for your own career

If your job appears in the table above, that is not a prediction that you will be unemployed in 2030. Every source on this page distinguishes between 'tasks an LLM can do' and 'jobs an employer will cut.' Even Goldman Sachs' most aggressive scenario projects that significant labor disruption from generative AI is most likely to begin showing up in productivity statistics after 2025 and to peak in the 2027–2030 window — and they explicitly note that displacement is typically offset by new task creation within the same occupations and by demand growth in adjacent ones.

The more actionable read is to look at the exposure number as a measure of how much of your week could be done faster with current LLMs, and to ask whether your employer captures that time savings as a headcount cut, a hiring slowdown, or a productivity-and-output increase. Industries with strong demand growth — software, healthcare, professional services — have historically reinvested productivity gains into more output rather than fewer workers. Industries with flat or shrinking demand are the ones where exposure converts most directly into headcount risk.

For a personal what-to-do-this-quarter checklist that maps to the same exposure scores, the calculator at AI tool cost vs. salary savings is a useful next step.

How to use this table to plan your 2026–2030 career

1. 1. Find your occupation's BLS SOC code

   Look up your role in the BLS Standard Occupational Classification on bls.gov/soc. The SOC code is the key that lets you cross-reference your role across Eloundou et al., McKinsey, BLS employment projections, and the WEF Future of Jobs Report. Without the code, you will end up comparing different definitions of the same job title.

2. 2. Pull the Eloundou et al. exposure score for that SOC code

   The OpenAI–UPenn paper publishes per-occupation exposure scores in its appendix on arXiv (2303.10130). Two numbers matter: the share of your tasks where an LLM alone can cut completion time by ≥50%, and the share where an LLM with additional tooling can. Treat the second as the 2030 number and the first as the 2026 number.

3. 3. Check the BLS employment projection for 2033

   BLS publishes 10-year employment projections per SOC code. If your role's projected growth is positive and above the all-occupations average, exposure is more likely to mean 'more output per worker' than 'fewer workers.' If the projection is negative, exposure compounds with structural decline.

4. 4. Map your weekly task mix to the exposure categories

   Pick a typical week and tag each block of work as high-exposure (text generation, summarization, code generation, structured data extraction) or low-exposure (in-person judgment, physical manipulation, regulated decisions). The share that lands in the high-exposure bucket is roughly the share of your week an LLM can already accelerate.

5. 5. Decide whether to capture the productivity gain or fight the headcount cut

   If your employer rewards higher output (sales, engineering, consulting), invest in AI tooling that lets you produce more per week. If your employer is in a flat-demand sector where the productivity gain converts to headcount cuts, the better hedge is to move toward the low-exposure tasks in your own role — judgment, client relationships, novel problems — before the cut cycle arrives.

Frequently asked questions