Someone tells me a robot is coming for my job roughly every second week. I do SEO and write for a living, so apparently I’m in the firing line. That’s why I went digging.
My take, for what it’s worth: I think AI will chew through tasks inside lots of jobs well before it makes any whole occupation vanish. So where’s that likely to show up first in Australia? What’s still anyone’s guess? And what can workers and bosses do in the meantime? That’s what the rest of this is about. Expect a rough sketch with blank patches rather than a forecast, because every industry, workplace and role will play out a bit differently.
What does it mean for AI to change a job?
“AI will change your job” sounds like one idea. It’s at least three, and they get muddled constantly.
First, automation. A tool does the whole task and nobody touches it. Self-serve checkouts are the everyday version, although anyone who’s been told about an unexpected item in the bagging area knows it only half works. Augmentation is gentler. The tool does a chunk of the job and you still make the call. Then there’s the one nobody mentions: the work gets reorganised around the tool. Say it writes every first draft. Your role slowly turns into editing, and the team shifts to match.
I’ll be saying “generative AI” a lot, so here’s what I mean. Basically it’s software that produces new stuff (text, images, code, summaries) when you give it a prompt, after being trained on a huge pile of existing material. Which is why it sounds so sure of itself. Sounding sure and being right are different things, and I’ll come back to that. You’ve probably heard of ChatGPT and Claude.
The Jobs and Skills Australia Gen AI Capacity Study is the best Australian source I came across. Its verdict was that generative AI is more likely to augment the way people work than to replace jobs through automation. That describes the technology today. It’s not a guarantee about ten years out.
Which tasks are most exposed to AI?
Headlines love job titles, but exposure is really about tasks. The stuff that’s easiest to hand to a machine is repetitive and made of words or numbers: drafting, summarising, sorting, ticking things off against a fixed set of rules.
Take Anthropic’s own exposure research. Computer programmers topped the list, with about 75% of their tasks exposed. Meanwhile about 30% of workers have no measurable exposure whatsoever. Think cooks, bartenders and lifeguards, whose work barely appears in AI usage data.
Exposed does not mean doomed. Say an accountant’s tasks are 60% exposed. That’s not a 60% chance of being sacked. It just means a good part of their week could look different, and what that turns into (fewer jobs, more output, or a less miserable Tuesday) is up to the boss, the customers and the rules of the industry.
Where might Australian workplaces feel the change first?
The federal government’s AI and employment report, published in July 2026, is the closest thing we have to hard Australian data. It found no evidence of broad labour market upheaval so far. Employment in the most AI-exposed fifth of occupations grew 5.6% between November 2022 and February 2026, against 9.5% for the least exposed fifth. The authors call their results suggestive, not definitive. Fair enough. A gap in growth doesn’t prove AI caused it.
One more detail from the report: clerical and administrative work, which sits high on the exposure list, was already shrinking before ChatGPT arrived. So admin work looks like a plausible early candidate for change, but some of that story predates AI.
Beyond admin, the areas I’d watch are customer service, finance and accounting, retail, health administration and professional services such as marketing, legal support and software. I’m guessing a little here. The honest answer is that nobody has a reliable league table.
Pace will depend on local factors, starting with who’s actually using this stuff. The National AI Centre’s adoption tracker had small and medium businesses at 41% in late 2024. I’ve seen bigger numbers quoted since, but I couldn’t pin down a reliable current one, so I’ll leave it there.
Business size looks like the biggest dividing line. One analysis of ABS figures put large businesses at about 35% adoption versus roughly 11% for small and micro ones in 2024-25. That’s roughly three to one. Then there’s whether good tools exist for a particular job, and the rules, which I’ll get to.
How AI could reshape everyday work
A few examples. They’re things AI tools might help with, not proven results everywhere, so please don’t quote me to your boss.
Marketing might get a rough first draft out of a tool and then rewrite most of it. A customer service team could have one sort enquiries by topic and urgency before a human replies. Finance people might ask for a summary of a 60-page document to decide whether it deserves a proper read. And a clinic’s admin staff might have it draft the routine letters nobody enjoys writing.
Every one of those examples assumes somebody checks the result. That step is where most of the risk sits. Jobs and Skills Australia’s case studies describe a similar pattern in health settings, where tools are meant to support clinical judgement with human oversight, not replace it.
None of it is necessarily common yet, either. In plenty of workplaces I suspect the story is still “someone tried it once, it made up a statistic, end of experiment”.
What could change for workers, and what may stay human-led?
My guess is that once drafting and summarising are cheap, the valuable bit is knowing what’s worth writing in the first place and spotting where the draft is wrong. Plus being the person whose name is on it. Judgement, plain talking and an actual feel for your customer should count for more, not less. Responsibilities may shift too, with junior tasks changing most. The government report found youth and graduate employment had mostly held up so far, but Anthropic’s exposure research found tentative evidence that job entry rates for 22 to 25 year olds into exposed occupations had dipped slightly. It’s early, and the two findings don’t neatly agree.
The least exposed work tends to be hands-on and care-based: electricians, plumbers, carpenters, truck drivers, cleaners and carers, according to the coverage of the report. Even then, the paperwork, quoting and scheduling around the job could change, though the job itself wouldn’t. And none of this is a verdict on your career in particular. It varies far too much by role for that.
How workers can prepare
There’s no single skill set that suits everyone. A bookkeeper and a nurse want very different things from these tools, and sometimes shouldn’t touch them at all. Some basics apply almost regardless of role.
Get to know the tools your role actually uses, and practise on something that doesn’t matter before you try it on anything with a client’s name on it. Check everything they give you. It can be confidently wrong, and convincing with it. Keep working on the skills around the tool (judgement, knowing your subject, looking after customers). And read your workplace’s policy on what you can paste in. Client information shouldn’t go into a public chatbot.
A course built around your own role will probably be worth more than a generic “AI skills” one. Your employer, your industry body or the nearest TAFE or university might have something. As for the bigger picture, Jobs and Skills Australia says adaptability will be critical, because new jobs will emerge while existing ones change.
What employers should consider before adopting AI
Buying the tool is the easy bit. Treating it as an IT purchase is probably where it goes wrong. Before anyone gets a login, I’d want these sorted.
A use policy that staff can actually follow, for a start. Then rules on privacy and data handling, meaning which customer and employee information goes into which tools. Consultation with staff, since people who are asked tend to use tools better than people who are told. Training. Quality checks, and a human responsible for the final output.
Checking is the bit that gets skipped. Among businesses using AI, only about half have even basic output checking before results reach customers, according to one analysis of National AI Centre data. It’s a write-up of the data, not the tracker itself, so go to the original before quoting it.
On the legal side, Australia has no standalone AI Act. So is it a free-for-all? No. The December 2025 National AI Plan confirmed that the government is sticking with existing laws (privacy, consumer protection, workplace relations and so on), and the earlier idea of mandatory guardrails got dropped.
For employers, that means the rules you already follow still apply. HWL Ebsworth says AI used in decision-making has to comply with the Fair Work Act and discrimination law, and new privacy rules on transparency around automated decision-making start in December 2026. Gadens ties those rules back to Robodebt, which probably tells you how seriously they’re meant to be taken. I’m no lawyer, so get proper advice before relying on any of it.
Looking beyond today’s tools
The tools I’ve been describing will probably look quaint in a few years. The harder question is how fast things move and who gets hit first. Is it a rush or a slow grind? Big employers, or small ones nimble enough to try things? Honestly, even the people paid to forecast this can’t agree. The government’s own report lists a wide spread of forecasts.
For a team day or conference, bringing in futurists and future trends speakers is one way to pull people’s heads out of this week’s software.
Where to find practical AI perspectives
Reports will only take you so far. At some point it helps to hear from someone who’s actually used this stuff, and who’ll cover the boring questions: what works, what doesn’t, what to ask a vendor before you sign. For your team, AI speakers are a decent place to start.
And whoever you book, ask them to separate what’s been measured from what’s just a prediction, and what they’re basing it on. Then ask whether any of it applies to your industry.
So, which jobs will change first?
My best guess is that it’ll be uneven. Some tasks in some roles will change soon. Plenty won’t for ages. Where you work will count for as much as the technology itself. If you want something to do this week, write down what you actually spend your time on, circle the repetitive word-and-number stuff, and try a tool on one of those tasks. Then come back to it in six months. The tools and the evidence will both have moved. Think I’ve got something wrong? Tell me, I’d rather know.
