The Question Everyone Is Asking
It's a reasonable thing to wonder. Generative AI can write code, draft legal briefs, analyze medical images, generate marketing copy, and answer customer service questions. If AI can do those things, what does that mean for the people who currently do those things for a living?
The short answer, based on the available evidence, is: replacement is rarer than feared, but disruption is real and unevenly distributed. The key is understanding which specific elements of your work are at risk, not whether your job title appears on a list.
What McKinsey and Goldman Sachs Actually Said
Two studies get cited constantly in AI-and-jobs discussions, usually without much nuance. The McKinsey Global Institute's 2023 report estimated that generative AI could automate work equivalent to 60-70% of employee time across the economy. The Goldman Sachs analysis estimated that 300 million full-time jobs globally are exposed to automation by AI.
Both numbers sound terrifying. Both require context. "Exposed to automation" is not the same as "will be automated." The McKinsey figure refers to specific task categories within jobs, not whole jobs. And the Goldman analysis spreads the exposure estimate across decades, not years, with significant uncertainty bands.
12%
of jobs in advanced economies are estimated to face high automation risk in the near term, defined as 5 years (OECD Employment Outlook, 2024)
The OECD's more conservative estimate, 12% of jobs at high near-term risk, is probably more useful for planning purposes. "High risk" means a majority of the tasks in that role can be automated with current or near-current technology, and there is economic pressure on employers to make that substitution.
Which Jobs Are Most at Risk
Risk is highest where three factors converge: the work is primarily cognitive (not physical), the outputs are well-defined and verifiable, and the volume of work justifies the investment in automation tooling. Based on those criteria, the highest-risk categories are:
- Data entry and processing: Document intake, form processing, and data transcription are heavily targeted by enterprise AI tools
- Tier-1 customer support: AI chatbots handle a growing percentage of support volume at companies across every industry
- Basic research and report generation: Roles that primarily synthesize existing information into standard formats face significant pressure
- Routine legal document review: Contract review and discovery support have seen major efficiency gains from LLM-based tools
- Basic financial analysis: Generating standard reports, budget variance analysis, and routine forecasting
Which Jobs Are Least at Risk
| Job Category | Risk Level | Why |
|---|---|---|
| Skilled trades (electrician, plumber) | Very low | Physical dexterity, variable environments |
| Mental health counseling | Very low | Human relationship is the product |
| Strategic leadership / CEO | Low | Judgment in novel high-stakes contexts |
| Nursing and direct patient care | Low | Physical care, trust, real-time adaptation |
| Teaching (K-12) | Low | Relationship, behavioral management, presence |
| Creative direction | Low-moderate | Taste, strategic judgment, client relationship |
What Skills Protect You
The skills that are hardest to automate share a pattern: they require either physical presence, real-time social judgment, or the kind of tacit expertise that comes from years of experience in variable environments and cannot easily be encoded in training data.
More practically, workers who learn to use AI tools well are in a much stronger position than those who resist them. The risk isn't AI replacing you. It's someone who uses AI replacing you. Across virtually every knowledge work field, AI fluency is becoming a baseline competency, much like knowing how to use a spreadsheet was by the 2000s.
3x
more likely to receive a callback: candidates who demonstrate AI tool fluency on their resume in AI-adjacent fields, compared to otherwise equivalent candidates (LinkedIn, 2025)
The Realistic Timeline
Automation rarely happens as fast as technologists predict or as slowly as incumbents hope. The pattern in past waves of automation, from agricultural mechanization to the ATM, is that displacement happens at the task level first, then the role level over years or decades, with significant variation by industry and geography.
The realistic timeline for most workers is not "AI replaces my job next year." It is "the mix of tasks I'm expected to do shifts over the next 3-7 years, with AI handling more routine components, and the expectation that I spend more of my time on higher-complexity, less codifiable work." That's not comfortable, but it's also not the apocalyptic scenario the headlines often imply.
The workers who will fare worst are those in high-risk roles who neither develop adjacent skills nor learn to use AI tools to increase their own output. The workers who will fare best are those who treat AI capability as a skill to develop, regardless of their field, and position themselves at the intersection of domain expertise and AI fluency.