Jobs that are harder for AI to replace (and how we decide)

The jobs hardest for AI to replace are, in our read, the hands-on and in-person ones: trades, care, health and personal service work. Notice we said harder — not "AI-proof," a promise nobody can honestly make. And here's the part worth your attention: the work we rate hardest to automate is also the work Australia is most short of, on the government's own 2025 data.

First, the honesty bit: this is our read, not JSA's

Jobs and Skills Australia's Occupation Shortage List measures demand. It says nothing about AI. The AI-resilience signal on Workworthy occupation pages is our editorial analysis — deliberately simple, fully disclosed, and worth exactly as much as its reasoning. Here it is.

AI — especially the current generation — most readily replaces routine, predictable, screen-based work. It struggles with work that is physical, in-person and non-routine: no language model rewires a switchboard, resets a broken arm or crawls under a house. So we band occupations by the kind of work they are, using the government's own occupation groupings:

• Harder for AI to replace: technicians and trades workers; community and personal service workers (carers, emergency services, hospitality). Physical, varied, human.
• Mixed AI exposure: everything in between — managers, professionals, drivers, labourers. Real exposure in parts (much professional deskwork is automatable), real resilience in others (surgeons, teachers, machine operators).
• More exposed to AI: clerical and administrative work — the most routine, screen-based band.

Three coarse bands, on purpose. A per-job percentage would look more scientific and be less honest — nobody can score 916 occupations' automation risk to a decimal place. We'd rather give you a transparent rule you can argue with than fake precision you can't.

The numbers: resilience and demand point the same way

Cross our bands with JSA's 2025 shortage ratings (released October 2025) and the pattern is stark:

The shortage column is JSA's; the banding is ours. Read it plainly: the economy is running short of exactly the kind of work AI can't easily do, while the most automatable band has almost no hiring difficulty at all. 68 occupations in our hardest-to-replace band have been in shortage every year from 2021 to 2025 — carpenters, electricians, plumbers, motor mechanics, welders, aircraft maintenance engineers and airconditioning mechanics among them. Browse any of them at /occupations to see the rating and our read side by side.

What this means at the kitchen table

If part of your worry is "will a machine take this job before my kid is 40," the honest answer is that nobody knows — but the kind of work is the best clue available. A kid starting a carpentry apprenticeship is training for physical, on-site, every-job-is-different work. A kid heading into routine administration is entering the band already showing the least demand. That's not a reason for panic; it's a reason to pick with eyes open.

What this doesn't tell you

Our bands are group-level, so they miss individual texture: some trades have automatable slices (prefabrication is changing parts of construction), and some clerical roles will evolve rather than vanish. AI also changes how resilient jobs are done — a mechanic diagnosing with AI tools is still a mechanic, but a different one. And no five-year data window predicts 2040. We'll update our read as the evidence moves, and we'll keep telling you it's a read. Anyone selling you "AI-proof careers, guaranteed" is lying about at least one of those three words.

FAQ

What jobs are safest from AI in Australia?

We won't say "safest" — nothing is certain. In our labelled editorial read, the harder to replace band is trades and care/personal-service work: physical, in-person, non-routine occupations. That band also has the highest official shortage rate (39% on the 2025 OSL), so demand data and automation logic currently point the same way.

Is "harder for AI to replace" an official government rating?

No. The shortage ratings are official (Jobs and Skills Australia). The AI-resilience bands are Workworthy's editorial analysis, based on the kind of work each occupation group does. We label the difference on every page, because trusting us should never require confusing the two.

Are trades really hard for AI to replace?

Harder — yes, in our view: on-site, physical, non-routine work is the profile automation handles worst. Not immune: tools, prefabrication and AI-assisted diagnostics will keep changing how trades work. "The job changes" and "the job disappears" are very different futures, and trades sit firmly in the first camp. That's our read.

Which jobs are most exposed to AI?

In our banding, clerical and administrative work — the most routine and screen-based group. Notably, only 2 of its 65 occupations are in national shortage on the 2025 list, the lowest of any band. The demand signal and the exposure signal agree.