The AI Job Impact Report 2026

A surgeon and a medical coder work in the same hospital, often on the same patient's case. Over the next decade, artificial intelligence will treat their careers very differently — and the reason isn't talent, seniority, or luck. It's the shape of the work. One job is judgment and hands in a room. The other is words and codes on a screen. AI has a great deal to say about the second and almost nothing to say about the first.

Most coverage of AI and work flattens that distinction into a single anxious question — will it take my job? — and answers with one of two stories. In one, AI erases half the workforce by the end of the decade. In the other, it's an overhyped autocomplete that will quietly fade. Both make good headlines. Neither survives contact with the data.

This report takes the more useful, more demanding position in between. AI is already reshaping work — but unevenly, predictably, and along lines you can see and measure. We scored 159 occupations on a single 0–100 scale, where higher means more insulated from AI. We call it the AI-Proof Score, and it draws on the strongest evidence available in 2026: real usage data from hundreds of millions of AI conversations, government and academic exposure indices, and employer hiring forecasts. What follows is not a prophecy. It's a map of where the ground is moving fastest — and where it is barely moving at all.

Key findings

1. Exposure is not the same as job loss — and the gap between them is enormous. AI is theoretically capable of about 94% of computer and mathematical tasks, yet today it's actually used on roughly a third of them (Anthropic Economic Index). Across the economy, capability runs years ahead of adoption. What a model can do and what it is asked to do are different numbers, and the distance between them is where careers are decided.

2. The most exposed jobs are clerical — not creative, not technical. The roles where AI already covers the most work are data entry, customer service, bookkeeping and administrative support: transactional desk work. Data Entry Clerk scores 16 of 100, the lowest in the report, and clerical roles cluster at the bottom of the scale. Our most-exposed group echoes the World Economic Forum's fastest-declining roles and the ILO's highest-exposure category almost line for line.

3. Nothing physical or deeply human ranks as exposed. Surgeons, electricians, plumbers and firefighters sit at the safe ceiling (94); nurses, caregivers and paramedics just below (90–92). Roughly 30% of workers are in occupations AI usage barely registers at all (Anthropic). Hands, bodies, and accountability someone has to stand behind remain the durable moats.

4. "Exposed" and "growing" are often the same job. Software developers, data scientists and UX designers score as meaningfully exposed — and are simultaneously among the fastest-growing roles in the world (World Economic Forum). Exposure measures how much of the work AI can do. It says nothing about how much of that work there will be.

5. Fluency with AI is the most controllable factor you have. In every field, using AI well moves your number more than almost anything else. AI-skilled workers earned a 56% wage premium in 2024 — double the year before (PwC) — and experienced users complete tasks measurably more successfully than beginners, by about 4–5 percentage points (Anthropic). The skill is learnable. The gap widens or closes by choice.

6. The labour market hasn't cracked — but one seam is showing. Through early 2026, workers in exposed occupations show no systematic rise in unemployment (Anthropic; Yale Budget Lab). The exception is the bottom rung: hiring of 22–25-year-olds into the most exposed roles is quietly slowing. The entry ladder is thinning before the whole structure does.

7. The net forecast is growth, not collapse. Employers expect AI and related shifts to displace 92 million roles by 2030 while creating 170 million — a net gain of 78 million jobs (World Economic Forum). The disruption is real. The apocalypse is not the base case.

Exposure is not destiny

A word about what these scores mean — and what they don't.

When researchers say a job is "exposed" to AI, they mean something precise and narrow: that a large share of its tasks are the kind a model can now perform or assist with. Exposure measures overlap, not outcome. It describes capability, not your employment.

That distinction carries the whole report, because capability is running well ahead of reality. The same Anthropic data that finds AI theoretically able to handle 94% of computer and mathematical tasks finds it actually used on about a third. The tools can already do far more than they are asked to. The gap between can and does is measured in years — years in which cost, regulation, trust, liability and plain institutional inertia all act as brakes.

This is also why the big numbers seem to argue with each other. Goldman Sachs put 300 million jobs "exposed" to automation worldwide; the World Economic Forum projects a net gain of 78 million. Both are correct, because they count different things — one counts tasks AI could touch, the other counts jobs employers expect to keep hiring for. After comparing the major studies, Yale's Budget Lab reached the cleanest summary of the field: the measures agree almost entirely on which jobs are exposed, and disagree mainly on how much. Direction is settled. Magnitude is the open question.

So a score here is information, not a verdict. A low number doesn't mean your job is ending — it means a meaningful part of what you do today is becoming cheap to automate, which is precisely the signal you'd want early, while there's still room to move toward the parts that aren't. The most exposed worker who sees it coming is in a stronger position than the "safe" one who assumes the ground will never shift. That is the spirit of everything that follows: not a countdown, but a map.

How to read the scale

Full methodology and sources are listed at the end of this report.

The rankings

We scored all 159 occupations on the AI-Proof Score, and the two ends of the scale are worth seeing first.

The ten most exposed occupations

Overwhelmingly clerical and transactional desk work — the roles where AI already covers the largest share of the day-to-day.

The ten most resilient occupations

At the safe end of the scale: hands, bodies, and accountability a person has to carry — work AI can describe but cannot do.

Find your own occupation

The full ranking of all 159 occupations is below — searchable and sortable, with the reasoning and the cited evidence behind every score, plus the field-by-field readouts that follow it.