What Jobs Will AI Replace in Australia? Exposure Scores for 357 Occupations

Jobs and Skills Australia — the federal government's labour market agency — scored the task content of every Australian occupation for how much of it generative AI could actually undertake. The result is the closest thing this country has to a definitive answer, and it is more reassuring than the headlines.

78% of Australian employment sits in the lowest exposure band. The average occupation scores 35 out of 100. The most exposed occupations in the country — Keyboard Operators and Telemarketers — score 81. The least exposed — Domestic Cleaners, Concreters, Handypersons, and Meat Boners and Slaughterers — score 10.

Two things matter before you read the tables:

Exposure is not the same as replacement. An exposure score describes what the technology could do with an occupation's tasks. It says nothing about whether any employer will act on it, when, or to whom. The federal Department of Employment and Workplace Relations reported in July 2026 that there was no evidence to date of broad labour market upheaval driven by AI in Australia — and that employment in the most-exposed occupations had still grown 5.6% since November 2022, just more slowly than the 9.5% growth in the least-exposed.

Most jobs score higher for augmentation than automation. Jobs and Skills Australia scored both. For 84% of Australian employment, the augmentation score — AI assisting the work — is meaningfully higher than the automation score. Bookkeepers score 69 for automation and 77 for augmentation. That gap is the whole story for most people reading this.

You can check your own occupation, adjust for how your actual day is spent, and see what you would be owed if a restructure ever came, with the Will AI Take My Job? exposure checker.

Scores are out of 100. "Automation" is how much of the occupation's task content generative AI could undertake; "augmentation" is how much it could assist with. Employment figures are Jobs and Skills Australia's 2025 estimates.

The pattern is not "white collar versus blue collar" — it is structured information handling. Every occupation on this list processes text, forms, records or scripted conversations to a repeatable pattern. That is precisely what large language models do.

The single most significant line is General Clerks: 297,907 Australians, the largest exposed occupation in the country by a wide margin. Add Accounting Clerks, Bookkeepers and Payroll Clerks and you have well over half a million people in high-exposure administrative work.

Notice also that six of these fifteen score higher for augmentation than automation — Tourism and Travel Advisers, Accounting Clerks, Financial Dealers, Bookkeepers, Debt Collectors and Credit and Loans Officers. For those roles the more likely near-term outcome is a changed job, not a deleted one.

The trades sit at the bottom for a specific reason, and it is worth understanding rather than taking on faith. The bottleneck is not intelligence — it is dexterity. Manipulating unfamiliar objects in unpredictable, unstructured physical spaces remains one of the hardest unsolved problems in robotics, and it has resisted decades of effort while software raced ahead.

For context on the other trades: Plumbers score 15 and Electricians 16. Both are firmly in the low band.

The honest caveat on the trades. This protection is real today and thinner over a longer horizon. Goldman Sachs puts humanoid robot shipments at roughly 250,000 by 2030 in its base case, and concentrated in structured industrial settings rather than on residential job sites. Morgan Stanley models around 40,000 jobs displaced by humanoids globally by 2030, rising to 8.4 million by 2040. The robot wave, on those numbers, lands mostly after this decade — but it is visible from here.

The assumption behind most AI-and-jobs coverage is that exposure and decline go together. In the Australian data they frequently do not. These occupations score well above average for AI exposure and are projected among the fastest-growing in the country over the next five years.

Software and Applications Programmers are the clearest example: 63 out of 100 for exposure, and projected to grow 15.7% by 2030 from an existing base of 186,066 workers. The Department of Employment and Workplace Relations found the same thing in the actual employment numbers — programmer employment was up 25% since November 2022, reaching 199,000.

Two forces are at work. Jobs and Skills Australia's projections already price in technology, so growth here is a forecast made with knowledge of the exposure. And in occupations where AI raises output per worker, cheaper output often expands demand faster than it displaces labour.

A few occupations people ask about most, with their scores and five-year employment projections:

Accountants land mid-table at 54 — but with an augmentation score of 73, the highest gap on this list. Solicitors score lower than most people expect at 34, because the bulk of legal work is judgment, negotiation and client relationships rather than document generation. Registered Nurses, at 30 with 363,766 people employed and 13.7% projected growth, are close to the safest large profession in the country.

These are occupation averages, and your own day may look nothing like the average for your job title. That is exactly what the exposure checker adjusts for — how much of your work is screen or phone based, whether it is relationship-driven, whether it is hands-on, and whether it could be delivered from another country.

There is a second displacement channel that occupation exposure scores do not capture, and in Australia it may be the more immediate one.

Work that can be done through a screen can usually also be done from another country. Australian restructures at Qantas, Telstra and Woolworths have moved screen-based roles offshore with AI cited alongside the decision. For an individual worker the practical question is not only "can a model do my tasks" but "can my tasks be performed somewhere cheaper, with AI making the handover easier".

The two compound, because they select for the same work: mediated, documented, deliverable at a distance. A role at the intersection carries both exposures at once, which is why the exposure checker scores offshoring on its own separate dial rather than folding it into the AI number.

Whatever the score for your occupation, the rules if a restructure arrives do not change because the reason was AI.

Your employer probably has to consult you first. Modern awards and enterprise agreements must contain a consultation term, triggered by a definite decision to introduce major change in production, program, organisation, structure or technology that is likely to significantly affect employees. Where it applies, your employer must notify affected employees, discuss the change and measures to avert or mitigate its adverse effects, and genuinely consider what you raise. It is not a veto — but skipping it has consequences. Check your position with the consultation duty checker.

"We automated it" does not automatically make a redundancy genuine. Under s.389 of the Fair Work Act, a dismissal is a genuine redundancy only if the employer no longer needs the job done by anyone, the employer complied with any consultation obligation in the award or agreement, and redeployment elsewhere in the business or an associated entity would not have been reasonable. If the job still exists under a new title, if consultation did not happen, or if redeployment was never properly considered, it may not be genuine. See the genuine redundancy challenge letter tool.

Redundancy pay does not depend on the reason. Under the National Employment Standards it runs from 4 weeks at 1–2 years of continuous service up to 16 weeks at 9–10 years, then drops to 12 weeks at 10 or more years — paid on top of notice and unused annual leave. Small businesses with fewer than 15 employees are generally exempt, and casuals generally have no entitlement. An award, agreement or contract can improve on the NES but cannot undercut it. Work out your figure with the redundancy pay calculator.

The unfair dismissal clock is short. An application to the Fair Work Commission must generally be lodged within 21 days of the dismissal taking effect. Missing it is usually fatal to the claim. Test your eligibility with the unfair dismissal checker.

The scores come from Jobs and Skills Australia's Generative AI Capacity Study, which assessed the task content of every ANZSCO occupation and produced separate automation and augmentation exposure scores. We use the published figures for 357 four-digit occupation groups, joined to the same agency's employment projections for 2025 to 2035. The data is published under CC BY 4.0.

What the scores genuinely cannot tell you:

• Whether your employer will act. Exposure measures technical task content, not commercial decisions. MIT researchers found in 2024 that only around 23% of the wages attached to exposure-eligible tasks were economically worth automating at then-current costs.
• Anything about you personally. These are occupation averages across every worker with that job title in Australia.
• What happens next. Forecasts in this field have a poor record. The widely-quoted 2013 estimate that 47% of US jobs were at high risk of computerisation, and the Australian "40% of jobs" figure derived from it, have not been borne out on anything like the predicted timeline.

An exposure score is a starting point for thinking, not a verdict on your career — and it should never be used by an employer to justify a redundancy. Whether a dismissal is a genuine redundancy is decided under s.389 of the Fair Work Act, on whether the job is still required, whether consultation happened and whether redeployment was reasonable. Never by a score.

Source: Jobs and Skills Australia, Generative AI Capacity Study (occupation data on AI exposure) and Employment Projections 2025–2035, both CC BY 4.0 © Commonwealth of Australia. Figures in this article are derived by FairWork Mate and are not endorsed by Jobs and Skills Australia or the Commonwealth.