AI is coming for office jobs, but not the ones you'd expect

The Australian government has done its first formal check on whether artificial intelligence will put people out of work — and found something counterintuitive. University graduates and women are overrepresented in jobs most at risk from AI, while tradespeople and care workers are among the safest The Guardian.

The report, AI and Employment in Australia, came from the Department of Employment and Workplace Relations and was published on 8 July DEWR. It's the first time Canberra has tried to measure this systematically, and the government says it will keep running this tracker.

What the report found

The headline finding you'll hear politicians repeat is this: AI hasn't caused widespread job losses in Australia yet. That's true — but it's worth hearing that carefully. It's a snapshot of now, not a promise about later.

The report looked at jobs and sorted them by how much of the work involves "routine cognitive tasks" — the kind of repetitive thinking that AI is good at. Think spreadsheets, form-filling, simple problem-solving. Jobs ranking highest on that list include telemarketers, call centre workers, office clerks, retail managers, software programmers, accountants, receptionists, and advertising and marketing professionals.

Jobs ranking lowest — the ones least likely to be automated — include tradespeople, aged care workers, carers, truck and forklift drivers, cleaners and gardeners. Workers in the high-risk jobs tend to be university-educated and female. Workers in the low-risk jobs tend to have less university education and more vocational training instead.

Why this matters

There's something odd hiding in these findings that deserves attention. We've spent the last decade telling young people that going to university is the best protection against losing your job to a machine. The data suggests the opposite might be true — at least for now. If AI is best at swallowing up the routine cognitive work inside accounting, marketing and clerical jobs, then "learn to code" and "go to uni" advice needs a second look.

Meanwhile, the jobs that seem most secure involve hands-on work: fixing things, caring for people, driving, building. These aren't jobs you can do through a computer.

What the government said

Employment Minister Amanda Rishworth framed the findings as a reason to keep supporting people rather than as a panic button. The government is "determined to ensure AI is harnessed to create good jobs, not threaten them" and will give Australians "the skills, training and pathways needed to adapt and benefit," she said The Guardian.

That's standard language for this kind of announcement — reassurance plus a promise of help. Whether the government actually delivers on that support is something worth checking against the next year's report.

The government also put something notable in the document: a quote from the head of AI company Anthropic predicting that AI could eliminate half of all entry-level office jobs and push unemployment to 10 to 20 per cent within one to five years. That's a genuinely alarming forecast, and including it in an official government report gives opposition and crossbench MPs something concrete to throw at the minister in parliament.

The timing question

The broader context here matters. The government was expected to announce updated plans on regulating AI in the week after this report came out. Releasing the jobs data first gives the government evidence to point to when it introduces its new rules, and lets it frame regulation as a measured response to actual risk rather than a knee-jerk reaction to lobbyists or scared voters.

The real thing to watch over the coming years isn't what this report says — it's whether the government actually keeps running this tracker properly. A one-off report is just a press release. A genuine tracker that keeps running is a tool that government can use to guide decisions on retraining money, visa settings for AI skills, and workplace rules about automation. Whether this becomes that serious tool or just a one-time news cycle is the question.