Pivotum Degree Risk Index · Fall 2026 Edition
Every profession in this index is scored the same way, against the same six questions, with the same weights. This page explains exactly how — because a score you can't interrogate is just an opinion with a number attached.
Read this before you trust anything else here. If the method doesn't convince you, the scores shouldn't either.
Each profession gets a risk score from 1 to 10, where 10 is most at risk from AI over the coming years.
The score is not a prediction that a job will disappear. Very few will. It's a measure of how much pressure AI is putting on that work — how much of it can be automated, how exposed the path into it is, and how much genuine protection the profession has.
A high score means "approach with your eyes open and choose your path carefully." It does not mean "avoid."
We score every profession on six factors. Three raise risk; three lower it. They're split evenly — 50% of the score comes from exposure, 50% from protection — so no profession is condemned or saved by a single number.
On the report, each factor appears as a plain question about your child's actual prospects. Here's what sits behind each one.
1. How much of this job can AI already do? — weight: 35% The largest single factor. What share of the day-to-day work is cognitive or routine — writing, analyzing, summarizing, drafting, standard production — that AI can already do or nearly do? The more of the job that's automatable, the higher the exposure.
2. How hard will it be to land that first job? — weight: 15% The factor most people miss. AI tends to automate the junior tasks first — the grunt work that entry-level roles were built from, and that people historically used to build experience. When those tasks vanish, the profession can stay healthy while the path into it narrows. For someone choosing a degree, this matters enormously: a safe career you can't break into is not a safe bet.
3. How exposed is the profession's high-value work over time? — folded into the factors above and the durability analysis We don't assign a speculative "future score" — nobody can predict the pace honestly. Instead we assess how durable each protection is (below), and we show how the score has moved over the past three years so you can see the direction of travel yourself.
4. Does it have to be done in person, with your hands? — weight: 15% Work requiring physical presence, dexterity, or a body in a specific place is far more insulated, because robotics lags software by a wide margin. A nurse turning a patient is safe in a way a remote analyst is not.
5. Does someone need a human they can trust and hold responsible? — weight: 15% Some roles require a human to be accountable — legally, for safety, or emotionally. A person who owns the outcome when something goes wrong; someone a frightened patient or anxious client needs to be human. This resists automation even when the underlying task is automatable.
6. Does the law require a licensed human? / How often does the job hit genuinely new situations? — weight: 10% + 10% Two durable protections. Licensing and regulation (medicine, law, nursing, engineering sign-off) mean AI can assist but cannot hold the credential — and this moves at the speed of legislation, not technology. And judgment under genuine novelty — situations that don't match any pattern, where being wrong is costly — is where AI remains weakest.
| Factor (reader-facing question) | Weight | Effect |
|---|---|---|
| How much of this job can AI already do? | 35% | raises risk |
| How hard will it be to land that first job? | 15% | raises risk |
| Does it have to be done in person, with your hands? | 15% | lowers risk |
| Does someone need a human they can trust and hold responsible? | 15% | lowers risk |
| Does the law require a licensed human? | 10% | lowers risk |
| How often does the job hit genuinely new, high-stakes situations? | 10% | lowers risk |
Automatability is the biggest lever, but the three protections together (50%) can outweigh it — which is why a job with lots of automatable tasks can still score low if presence, trust, and licensing hold. That balance is deliberate.
The weights don't change between editions. They're a fixed ruler. If we re-tuned them every six months, a score moving from 6 to 7 might just mean we'd changed the measurement — not that the world had changed. Because the ruler stays fixed, a score that moves tells you something real. We treat any change to the methodology as a rare, clearly-documented event.
"Lawyer" is not one job. A document-review associate and a trial litigator face completely different levels of AI exposure. So do a bookkeeper and an audit partner, a junior developer and a systems architect, a bedside nurse and a telehealth triage nurse.
Averaging them into one number for "law" or "nursing" hides the single most useful thing you could know. So wherever it matters, we score the tracks within a profession separately. This is usually where the real decision lives.
AI is moving fast enough that a static score would mislead. Two things address this:
Three-year history. For this first edition, we retrospectively scored each profession for 2023 and 2025, applying today's methodology to what was actually known in those years. This shows the velocity — how fast a score is moving — which often matters more than where it sits today. These earlier scores are reconstructed, not archived from past editions, and we say so on every profile.
Durability analysis. Rather than predict future scores, we assess how long each protection is likely to hold. Licensing and human-accountability are durable — they're rooted in law and society. Physical protection is strong today but is the one physical AI directly targets. Judgment is a real protection now but the exact frontier AI is advancing on. We tell you which shelters are solid and which are on borrowed time, and let you draw your own conclusions.
This is the most important thing to understand about every number here, and it is easy to miss.
We score exposure to what AI can already do. We do not score how much of it any particular employer has actually deployed.
Adoption lags capability, often by years, and it lags unevenly. Large, well-funded organizations with technical staff move first. Small practices, rural providers, public-sector employers, regional firms and anyone running on thin margins move much later — sometimes a decade later.
This means the same job title can carry very different day-to-day exposure depending on where it's done. A nurse at a large teaching hospital using ambient documentation is already working in the world these scores describe. A nurse at a small rural facility may not encounter any of it for several years.
We score the frontier deliberately, because that's what makes this a leading indicator rather than a lagging one. A score that reflected average deployment would tell you where things already are, which is far less useful for a decision made years in advance.
But read the lag correctly: it is a buffer, not a shelter. It buys time — potentially quite a lot of it, in the right setting. It does not change the direction of travel, and it is not something to build a thirty-year career plan around.
It is: a consistent, transparent framework applied evenly across professions, built to help you ask better questions and see distinctions the headlines miss.
It isn't: a guarantee, a prediction, or financial advice. These scores reflect informed analysis and judgment about a fast-moving situation. They're a strong starting point for a decision — not the decision itself. The person choosing a path should weigh this alongside their own interests, aptitudes, and circumstances.
We update every six months, because the honest truth about AI and careers is that anyone claiming a permanent answer is selling one.
Questions about the methodology are welcome — reasoned challenges are how the ruler stays honest, and how it improves between editions.
Yes, AI helped write this. We build AI systems for a living, so it would be a strange thing to hide.
Ours: the framework, the weights, 158 scoring judgments. The machine's: turning that into sentences that don't read like a spreadsheet with opinions.
Every figure is sourced. Every score publishes its arithmetic. And when we're wrong — we already have been, twice — we say so.