What to do about it
Four tests, applied to any role you are considering — plus the three ways to be wrong.
The work needs a body, in a place that keeps changing
Robots reach about 13% of US work hours today against 44% for software agents, because machines lack fine motor control and real-time physical judgement.
mckinsey-agentsA person is legally or morally answerable for the result
Indivisible roles survive: every school still needs a principal, every hospital a medical director, and nobody insures a model.
mckinsey-workforceIt depends on a relationship with a person
New tasks added to AI-exposed roles since 2022 are 2.5× more likely to depend on empathy, presence, judgement, creativity or leadership.
pwc-barometerThe role is one job, not a pile of tasks
More than 70% of today’s skills apply to both automatable and non-automatable work — the seat stays, the mix changes.
mckinsey-agentsNote: the 0–3 scoring shown on each job profile is an editorial reading of its task profile, not a measurement. The evidence above is sourced.
Test 1 — Does it need a body in an unpredictable place? If conditions change daily and a mistake injures someone, it is durable for a long while yet. The strongest evidence is the ability ranking: stamina, strength, reaction time, spatial orientation and problem sensitivity in settings where conditions change.Resume Now 2026
Test 2 — Is a human legally or morally on the hook? Accountability roles grow as automation grows, because someone must own the output. Regulation makes it explicit: the builder cannot be the auditor.Axipro 2026
Test 3 — Would a customer be upset to learn no human was involved? Care, teaching, therapy, negotiation. That discomfort is a moat, and the data agrees — new tasks in AI-exposed roles lean 2.5× harder on empathy, presence and judgement.PwC 2026
Test 4 — Is the job a bundle with an indivisible core? The safest jobs are the ones the org chart requires the seat for, not just the tasks.McKinsey 2025
Then act on the one finding with a clear instruction in it. AI-exposed entry-level roles that were redesigned to demand senior skills grew 35% since 2019; those that were not shrank 10%.PwC 2026 The move is not to avoid AI-exposed fields. It is to enter them at the expert end, which now means demonstrating judgement years before the career ladder would once have handed it to you.
Anti-test: if the job is “produce a first draft of a document”, it is already gone — not the profession, the rung. One in seven US workers has a direct skills pathway from a shrinking occupation into a growing one.McKinsey 2025 The jobs exist. The bridge is the bottleneck.
None of this is a forecast of which future arrives. The World Economic Forum’s own 2026 scenario work sets out four plausible 2030s depending on how fast capability advances and how ready the workforce is — from Supercharged Progress to Stalled Progress — and declines to pick one.WEF 2026 The tests above are useful in all four, which is the most any advice can honestly claim.
Three ways to be wrong
What would have to change for this advice to be bad advice.
The instruction
Do not avoid AI-exposed fields. Enter them at the expert end.
The rungs that vanished were the ones where the work was the first draft. What replaced them demands judgement, ownership and coordination years earlier than the old ladder would have. "Seniorise yourself early" is the whole actionable content of this brief, and it is the only recommendation here that a dataset actually supports.
The bottleneck is not the jobs — it is the bridge. About one in seven US workers has a direct skills pathway from a shrinking occupation into a growing one.McKinsey 2025