The board asked which half of the workforce is exposed, and the old assumptions don't hold
A People Analytics Lead walks into a planning meeting expecting to talk about warehouse automation and gets a different question: which of our finance analysts, contract reviewers, and marketing coordinators are exposed to generative AI. For two decades, "automation risk" meant robotics on a production line. That assumption is now out of date, and building a workforce plan on it will misdirect the analysis. This article lays out why exposure has shifted toward cognitive-routine, white-collar task bundles, where blue-collar exposure still concentrates, and how to read both without turning a task-level input into a verdict about anyone's job.
Why the automation story used to point at blue-collar work
Industrial-era automation displaced physically routine, repeatable motion: assembly, packaging, material handling. That pattern was real, and it shaped a generation of workforce planning that treated "blue-collar" and "automatable" as nearly synonymous. Generative AI does not follow the same task profile. Large language models are strongest at language, synthesis, drafting, and pattern-matching across text and data — cognitive work, not physical work. McKinsey's 2023 analysis estimated that 60–70% of employees' time is spent on activities that could theoretically be automated with current and emerging technology, up from roughly 50% in pre-generative-AI estimates, and the increase is concentrated in tasks involving natural language and knowledge work rather than physical manipulation.
That single shift explains most of what follows: the occupations now showing the highest exposure scores are not the ones the last automation wave targeted.
Where cognitive-routine exposure concentrates today
Brookings' 2024 analysis of generative AI's labor-market effects found that the highest-exposure sectors are STEM, business and finance, engineering, and law — fields built on drafting, analysis, and structured judgment applied to text and numbers. The same analysis estimates that roughly 12.9 million workers, about a third of the people employed in those occupations, are highly exposed. The Federal Reserve's 2026 work on AI adoption found a similar pattern from the usage side: measured work-related AI adoption runs roughly 5% to 40% depending on occupation, with professional services and financial-sector roles showing the highest and fastest-growing adoption. Anthropic's Economic Index reported that the share of occupations where its model was used for at least a quarter of associated tasks rose from about 36% in early 2025 to a pooled 49% — again concentrated in knowledge-work task types.
This is the pattern our cognitive routine tasks and AI explainer covers in depth: it is not "white-collar jobs" as a category that is exposed, it is the cognitive-routine task bundles that many white-collar roles happen to be built from — first-draft writing, standard analysis, templated research, routine coding. A role with a high proportion of those tasks scores higher on exposure regardless of its collar color; a role with a low proportion scores lower for the same reason.
Where blue-collar exposure still shows up — and where it doesn't
None of this means physical-routine work is exempt from the exposure conversation, and framing it that way would be its own analytical error. Physically routine tasks remain one of the four dimensions in a proper exposure assessment, and automation pressure on structured physical tasks — inspection, sorting, certain forms of quality control — has not gone away; it has simply been joined by a second, faster-moving pressure on cognitive-routine tasks. What has changed is which dimension is moving fastest and where the newest capability gains are landing. McKinsey's research on employment mix through 2030 projects growth concentrated in healthcare, STEM, and managerial occupations, and contraction in customer service, office support, and food service — a mix of white-collar and blue-collar categories on both sides of that ledger, which is the more accurate way to read it than a simple collar-color split.
The four-dimension rubric that makes the comparison rigorous
Comparing white-collar and blue-collar exposure honestly requires scoring the same four dimensions for every occupation, not applying a different lens to each:
- Cognitive routine (0–100) — drafting, standard analysis, templated research, rule-based decision-making.
- Physical routine (0–100) — repeatable manual tasks, structured inspection, material handling.
- Social/judgment (0–100) — negotiation, mentorship, contextual decision-making, stakeholder management.
- Creative (0–100) — original ideation, novel problem framing, design judgment.
WorkforceAnalysis maps an organization's role list to ONET occupations and scores each occupation's constituent tasks against these four dimensions, then weights the result by each task's ONET-documented importance to the role rather than treating every listed task as equally significant. A financial analyst role, for example, might score high on cognitive routine (structured reporting, variance analysis) and low on physical routine, while a facilities technician role might score the inverse. Neither score is a verdict — it is a structured description of where the role's task time is concentrated, which is the input a workforce planning conversation actually needs.
Worked example (illustrative, round inputs): suppose a contract-review specialist role has task-importance weights of 55% cognitive-routine work, 10% physical-routine, 20% social/judgment, and 15% creative, with raw dimension scores of 80, 15, 45, and 30 respectively. The importance-weighted composite is (0.55 × 80) + (0.10 × 15) + (0.20 × 45) + (0.15 × 30) = 44 + 1.5 + 9 + 4.5 = 59. That composite score routes the role to a Monitor or Review tier, not to any conclusion about the person in it — it flags where a redeployment conversation or a task-redesign conversation might start.
Reading the comparison without turning it into a verdict
The most common mistake in this comparison is collapsing "high exposure" into "will be automated" or "low exposure" into "safe." Neither is accurate, and neither is how the underlying research frames the finding. WEF's Future of Jobs Report 2025 projects a churn pattern, not a one-directional loss: 170 million jobs created and 92 million displaced globally by 2030, a churn rate of 22% of today's total employment, netting to roughly 78 million new jobs — alongside a finding that 39% of core skills are expected to be transformed or become outdated in the same period. That is a picture of skills shifting inside and across roles, which is exactly why WorkforceAnalysis reports exposure using Monitor, Review, and Redeployment Candidate tiers instead of pass/fail labels — the tier tells a workforce planning team where to look, and the O*NET-grounded redeployment options tell them what adjacent occupations share the skills already present in a high-exposure role.
Our generative AI exposure for knowledge workers piece goes deeper on why the knowledge-work pattern differs from the industrial-automation pattern, and which jobs are most exposed to AI breaks the occupation list down further. If you want the mechanics of how a raw O*NET occupation code becomes a scored, tiered result, AI exposure by occupation, explained walks through the pipeline end to end.
What this means for a mixed workforce
Most mid-market organizations run a mixed workforce — operations staff, field technicians, and knowledge workers sitting inside the same department structure, often the same P&L line. A useful exposure assessment has to score all of it on the same four dimensions, because a department heatmap that only covers desk roles will miss where physical-routine exposure is concentrated, and one that only covers frontline roles will miss the cognitive-routine shift that is currently moving faster. The goal of the exercise is not to rank collar colors against each other; it is to give a workforce planning team a defensible, occupation-level starting point for a conversation about training, redeployment, and task redesign — grounded in the same rubric across every role in the org chart. See our pricing page for how a department-wide assessment is scoped, and the Workforce AI Readiness Assessment Guide for a structured way to prepare your role list before you run one.
This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.
If you want this kind of methodology breakdown in your inbox as we publish new occupation-level research, subscribe to the WorkforceAnalysis newsletter below.