The best predictor of AI automation risk is not industry but task type. Jobs built on routine, rule-governed, predictable work — data entry, transcription, basic customer service — are most exposed. Jobs requiring physical unpredictability, deep human trust, or creative judgment are considerably harder to automate.

What Makes a Job More (or Less) Automatable

Research by Oxford economists Carl Frey and Michael Osborne identified three characteristics that make automation hard: complex physical perception and manipulation, creative problem-solving, and social intelligence. Where all three are low — form processing, data entry, telemarketing — AI can substitute well. Where one or more is high, substitution is difficult.

Generative AI has raised the stakes since those studies. McKinsey’s 2023 analysis found that the share of tasks in “applying expertise” roles that are technically automatable jumped 34 percentage points between 2017 and 2023, as language models gained the ability to draft documents, synthesize research, and generate code.

Jobs at High Risk

The Oxford study gave telemarketers a 99% computerization probability, data entry clerks 97%, and tax preparers 94%. Broader categories with high exposure include:

  • Office administration and data processing — form handling, scheduling, basic bookkeeping
  • Customer service (tier-1 support) — FAQ-answering, account queries, simple complaints
  • Entry-level legal and financial tasks — document review, contract summarization, basic research
  • Content transcription and basic translation — audio-to-text, templated writing

Goldman Sachs estimated in 2023 that 44% of legal research tasks and 46% of U.S. office-support roles have high automation potential. Neither figure means those jobs disappear — it means a significant share of the tasks within them can be done by AI, reshaping what the remaining human role looks like.

Jobs That Are Harder to Replace

Several categories resist automation for structural reasons:

  • Skilled trades — electricians, plumbers, HVAC technicians work in unpredictable physical environments where every job differs and physical judgment is essential.
  • Healthcare (hands-on) — nurses, physical therapists, and surgeons require direct contact, real-time physical assessment, and patient trust that AI cannot substitute.
  • Social and emotional roles — social workers, therapists, counselors, and teachers navigate complex human context and relationship-building that models cannot replicate.
  • Senior leadership and strategy — decisions involving ambiguous trade-offs, organizational politics, and long-term accountability require judgment AI can advise on but cannot hold.

The World Economic Forum’s 2025 Future of Jobs Report projects 92 million roles displaced by 2030 but 170 million new roles created — a net gain of 78 million, concentrated in green energy, care work, and technology.

The Augmentation Reality

AI more often changes jobs than eliminates them outright. The historical pattern from decades of automation research: automation eliminates tasks within roles, forces role redefinition, and creates new roles — but those new roles are not guaranteed, and the transition imposes real costs on workers whose skills do not transfer. Workers whose entire role consists of automatable tasks are most at risk; workers who use AI to handle routine tasks while doing higher-judgment work are typically more secure.

The practical takeaway: the question to ask is not “will AI replace my job” but “what portion of my tasks are routine and predictable, and what would remain if those were handled by AI?” See our guide on how to use AI to upskill for concrete next steps.

In the News

This pattern is showing up in labor data. A recent report found that AI was cited in 31% of U.S. job cuts for the fourth consecutive month — concentrated in back-office, content, and customer-service roles that match the high-exposure categories above.

FAQ

Does AI automation mean mass unemployment?
Major forecasts (WEF 2025, McKinsey 2023) project more jobs created than destroyed in aggregate, but the displaced and the newly-employed are often different people in different sectors, making the transition costly for many workers even if the net figure is positive.

Which skills protect workers most?
Adaptability, social reasoning, physical dexterity in unpredictable settings, and the ability to direct and verify AI output rather than compete with it.

How quickly is automation actually happening?
Historically, full technology adoption takes 10–20 years even after something is technically feasible. Generative AI is moving faster than previous cycles in white-collar work, but physical and relationship-intensive roles remain constrained by non-technical barriers.

Is the 99% automation risk for telemarketers a reliable figure?
It comes from a 2013 Oxford study measuring task characteristics against the technology of that time. Actual displacement is slower than pure technical feasibility — regulatory, social, and cost factors all slow adoption. Read it as “highly exposed” rather than “disappears by a given date.”