Most advice on AI job risk is anecdotal. Five structural factors decide whether your role gets compressed first or holds value as automation accelerates.
Most advice about whether AI will replace your job is anecdotal. Someone read a headline, looked at their own workflow, and made a personal call. The structural answer is different. Five factors decide whether a role gets compressed first or holds its value as automation accelerates, and you can read them off your own work in about ten minutes.
We rank 1,591 roles against these five factors in our AI career risk rankings. The full ranked list of the 100 most-exposed roles lives at the 100 jobs most exposed to AI. This article explains how to read the same factors against your own role so the ranking makes sense for your situation.
Factor 1: Routine task share
What percentage of your weekly workflow is repeatable, rule-bound, or templated? Drafting a standard report, processing an invoice, summarising a call, writing a routine email, formatting a deck, building the same kind of model you built last quarter. The higher the share, the larger the layer of your job that current AI tooling can already substitute. A role with 60-70% routine task share faces direct compression at the production layer; a role at 20-30% absorbs AI as augmentation rather than substitution.
Be honest about what is genuinely judgment work versus what just feels like it. A senior analyst who insists their work is non-routine but spends most of the week building variations of the same model is not safe. The exposure score on each role page in the AI career risk index breaks routine, strategic, and relational shares so you can see the split per role rather than guess.
Factor 2: Output substitutability
If a colleague handed an AI tool the same brief and the same data, would the output be acceptable? For copywriting, basic legal research, standard market analysis, code stubs, customer service triage, and routine creative production, the answer is now yes for a substantial share of work. Acceptable is a low bar, but the bar that matters: the moment AI output crosses it, the production layer of the role starts pricing down.
Roles that survive are those where the output requires a sequence of judgment calls AI cannot make consistently: complex deal architecture, partner-tier client trust, regulated advisory, surgical procedure, courtroom advocacy, original research direction. These are not safer because AI is technically incapable; they are safer because the buyer or regulator demands a named human signing the work.
Factor 3: Named accountability
Does the work require a licensed human to sign for it? Lawyers, doctors, chartered accountants, professional engineers, architects, and senior executives carry statutory or fiduciary accountability that AI cannot legally absorb. The licensure layer is the moat. Even when AI does most of the work behind the scenes, the named human stays in the value chain because the rules require it.
Roles without licensure or named accountability face a steeper curve. Marketing coordinators, junior analysts, paralegals, customer service representatives, and content producers all do work that flows through to a manager or system without their name on the final output. That structural feature makes them substitutable in a way licensed work is not.
Factor 4: Stakeholder relationship density
How much of the job depends on sustained trust with specific humans (clients, patients, students, regulators, executives)? Enterprise sales reps holding a multi-year customer relationship, family physicians, partner-track lawyers, executive coaches, and senior advisors all carry relationship density that resets to zero if you swap them out. This is structural protection because the relationship itself is the product, not a deliverable AI can substitute for.
Transactional or low-touch roles do not have this protection. SDR work, support tier-one, routine compliance reviews, vendor-managed services, and outsourced operations all run on interchangeable touchpoints. AI agents replicate interchangeable touchpoints at near-zero marginal cost.
Factor 5: Physical presence requirement
Does the work happen in person, on-site, with hands on equipment, animals, patients, or property? Surgeons, nurses, electricians, plumbers, carpenters, paramedics, dentists, field engineers, and skilled trades all have presence-bound work that AI cannot perform remotely or at machine scale. Even when AI augments their tooling (imaging, diagnostics, scheduling), the presence requirement remains.
Knowledge work without a presence component is more exposed by definition. Office-based analytical, administrative, communications, and routine creative roles all share this vulnerability. The work happens on a screen and the output ships through a screen, which is the same surface AI tooling now operates on.
Reading your own role against the five factors
Score yourself honestly across all five. A role with high routine share, high output substitutability, no licensure, no relationship density, and no presence requirement is in the high-risk band. A role with low routine share, low substitutability, statutory licensure, dense relationships, and a presence requirement is in the durable cohort. Most roles sit somewhere in between, and the practical question is whether the trajectory of your work over the next five years moves you toward the durable end or the exposed end.
If you want a deterministic score on the same five factors, find your role in the role index and read the structural reading on its page. If you want to see which 100 roles score highest on exposure across the whole index, see the most-exposed rankings. The ranking does the calculation; this article gives you the lens to read it.
What to do if your role is at risk
Three practical moves. First, target the durable layer in your current discipline: judgment work, accountability-bearing roles, system design, client trust, and the work AI cannot legally or practically own. Second, pursue licensure or specialism in regulated areas where the rules require a named human. Third, pivot toward an adjacent role on the AI-safe jobs list where the structural moats are strongest. The role-page pivot maps surface the realistic next steps from your current position.
Avoid the trap of becoming more efficient at the same work. Faster production output is the part of the job that AI absorbs first. The roles that pull away in the AI economy are the ones whose AI tooling makes them ten people wide rather than one person faster.
Run your structural assessment
The AI Career Index is free to take. Get your structural snapshot in 10 minutes, including your durability score, archetype, and projection.
Take the Free Assessment