A board asks a question most people leaders can't answer yet
A People Analytics Lead is three weeks from a board meeting. The company's private equity owner has asked a version of the same question every portfolio company is getting this year: which of our roles are exposed to AI, and what are we doing about it? The honest answer, right now, is a shrug dressed up in a slide. There's no task-level inventory, no defensible scoring method, no way to say which departments carry the most exposure versus which ones simply feel like they do.
This is not a failure of effort. It's a gap in tooling. Most organizations have workforce data, headcount plans, and org charts — but nothing that connects a specific role, in a specific department, to the actual tasks that compose it, and then scores those tasks against what generative AI can plausibly do today. That gap is what this article is about: how AI is genuinely changing work, what the credible research actually says (and doesn't say), and how a people leader builds a grounded, task-level view instead of guessing under board pressure.
What "AI is changing work" actually means at the task level
The phrase "AI is changing the future of work" gets used so often it has become almost meaningless. The useful version of the claim is narrower and more mechanical: jobs are bundles of tasks, and generative AI changes the economics of specific tasks, not entire job titles.
The U.S. Department of Labor's ONET database is the standard reference for this kind of analysis. It currently documents 1,016 occupational titles covering 923 data-level occupations and more than 55,000 job titles, described across roughly 277 descriptors that are updated on a regular cycle, with a primary annual update typically landing in the third quarter. The Bureau of Labor Statistics' 2018 Standard Occupational Classification system, which ONET aligns to, organizes the U.S. labor market into 867 detailed occupations, 459 broad occupations, 98 minor groups, and 23 major groups. That granularity matters, because it's the difference between saying "customer service is at risk" and being able to say which specific tasks inside a customer service role — say, routing a ticket versus de-escalating an angry client — carry different exposure profiles.
This site incorporates information from O*NET. Used under the CC BY 4.0 license. O*NET is a trademark of USDOL/ETA.
Anthropic's Economic Index, drawing on real usage patterns, found that 36% of occupations had at least a quarter of their tasks touched by Claude usage in early 2025, rising to a pooled 49% later in the year. A separate Anthropic analysis found that 68% of that usage fell into tasks assessed as fully feasible for AI augmentation or automation, while only 3% were judged not feasible — a useful reminder that most AI usage today assists with parts of a job rather than replacing the job outright.
What the research says — and doesn't say
The macro research is more consistent in direction than in scale, which is exactly why it should inform planning without dictating specific personnel decisions.
Goldman Sachs estimated in 2023 that generative AI could expose the equivalent of 300 million full-time jobs globally. McKinsey's 2023 analysis of generative AI's economic potential found that 60–70% of employees' work hours could theoretically be automated with current and anticipated technology, up from roughly 50% in pre-generative-AI estimates. These are exposure estimates — theoretical technical potential — not predictions of headcount reduction, and neither firm frames them that way.
The World Economic Forum's Future of Jobs Report 2025 offers the most widely cited jobs-churn numbers: 170 million jobs are projected to be created and 92 million displaced by 2030, a churn of 22% of today's total employment, netting to roughly 78 million new jobs on balance. The same report estimates that 39% of workers' core skills will be transformed or become outdated by 2030, and that 63% of employers cite skill gaps as the biggest barrier to workforce transformation.
Brookings' 2024 research on generative AI and the American worker found that the highest-exposure occupational categories are concentrated in STEM, business and finance, engineering, and law, and separately estimated that roughly 12.9 million workers — about a third of those in the occupations studied — are highly exposed. McKinsey's Global Institute projects that employment mix will shift toward healthcare, STEM, and managerial occupations, and away from customer service, office support, and food service roles by 2030.
Read together, this research supports a consistent, non-alarmist conclusion: exposure is real, uneven across occupations, and substantial in scale — but it is a distribution of tasks being reshaped, not a verdict on which people keep their jobs. That distinction is the entire premise behind potential impact of AI on US occupations, which goes deeper into the occupation-level data this article only summarizes.
Which jobs are most exposed, and why that's the wrong first question
"Which jobs are most exposed to AI" is the question every executive asks first, and it's worth answering carefully — see our dedicated breakdown for the occupation-level view. But asked in isolation, it tempts people leaders toward a binary that the data doesn't support: safe versus unsafe, kept versus cut.
A more defensible framing separates exposure from outcome. Exposure describes how much of a role's task content overlaps with what current AI systems handle well. Outcome — whether a role is redesigned, reduced, redeployed, or left untouched — is a management decision informed by exposure, budget, growth plans, and dozens of factors no algorithm can see. This is why our methodology at WorkforceAnalysis never labels a role "safe" or "automated." Every scored role lands in one of three action categories — Monitor, Review, or Redeployment Candidate — because the tool's job is to produce a defensible input to a human decision, not to make the decision itself.
Here's how that scoring works, using a simplified worked example. Say a role's tasks are weighted by ONET task-importance data and scored across four dimensions, each on a 0–100 scale: cognitive routine (how repetitive and rule-based the reasoning is), physical routine (how much of the work is manual and repeatable), social/judgment (how much depends on negotiation, empathy, or accountability), and creative (how much requires original synthesis). A role where cognitive-routine tasks score 80, physical-routine scores 10, social/judgment scores 55, and creative scores 30 — weighted by how much time ONET data says each task actually consumes — produces a composite exposure score and a corresponding action tier. Change the weighting toward social/judgment-heavy tasks, and the same job title can land in a different tier entirely. That sensitivity is the point: two people with the same title can carry different exposure profiles depending on how their actual task mix breaks down.
The reskilling gap: what organizations are (and aren't) doing about it
The WEF's Future of Jobs Report 2025 also quantifies the readiness gap behind all of this. Of roughly 100 workers who will need training as roles evolve, it estimates that 59 need some form of training, of whom 29 are expected to be upskilled in their current role, 19 reskilled or redeployed internally into a different role, and 11 unlikely to receive adequate training at all. That last group — the 11 — is the population every workforce transformation plan should be built around, because it represents real people whose skill gap is visible in the data but unaddressed in practice.
Gartner's 2024 research suggests most organizations aren't structurally ready to close that gap. It found that 86% of HR leaders have not implemented strategic workforce planning, and that 66% say their workforce planning is limited to basic headcount planning and struggle to demonstrate its ROI to the business. Lightcast's 2025 analysis of skill velocity adds urgency: 32% of the skills required in an average job changed between 2021 and 2024, and a quarter of jobs saw 75% of their required skills turn over in that window. Skills are moving faster than most planning cycles can track manually.
How to build a defensible, task-level view of your own workforce
None of the research above tells a specific company what to do with a specific department. That's the gap between macro research and operational planning — and it's the reason task-level, organization-specific analysis matters more than another market-wide statistic. A defensible internal view needs three things: a mapping of your actual roles to O*NET occupational data, a transparent scoring rubric applied consistently across every role, and an output that routes into action tiers rather than pronouncements.
That's the workflow WorkforceAnalysis is built around — self-serve, O*NET-grounded, and priced for mid-market teams rather than structured as a six-figure consulting engagement. If you want a lower-lift starting point before running a full assessment, the Workforce AI Readiness Assessment Guide in our store walks through the same rubric on paper. You can also review pricing for the full platform when you're ready to move from framework to a live, department-by-department heatmap of your own organization.
Where to start this quarter
The future of work isn't arriving as a single event. It's arriving task by task, quarter by quarter, inside job titles that mostly still exist and mostly still matter. The organizations that handle this well aren't the ones with the scariest slide — they're the ones that can show their board a methodology, not a mood. Start by reading the occupation-level research linked above, get familiar with the four-dimension exposure model, and treat every number in this space, including ours, as an input worth interrogating rather than a verdict to accept.
If you'd like grounded research like this delivered as we publish it — without the hype cycle — subscribe to our newsletter from the blog.