The Science Is In:
AI Adopters Aren't Cutting Jobs. They're Creating Them.

There is a particular kind of satisfaction that comes from watching a large body of received wisdom collapse under the weight of actual evidence. For the past several years, the dominant narrative around AI and employment has been one of inexorable replacement — a slow-motion Terminator scenario in which the machines gradually absorb the work, the humans gradually run out of things to do, and economists write increasingly anxious papers about what happens next. It is a compelling story. It has the advantage of being plausible, frightening, and easily illustrated with graphics of robots at desks.

It also, according to new research from CSIRO, does not appear to be what is actually happening in Australian workplaces.

Dr Claire Mason and her colleagues at CSIRO's Workforce and Productivity Research team analysed the hiring patterns of more than 4,000 Australian firms over three years, from 2020 to 2023. They identified which companies had adopted AI — based on signals in their job postings — and then compared what those companies did next with what the non-adopters did. The results are, by the standards of social science research, rather startling: firms that had adopted AI posted 36 percent more non-AI job advertisements over time than comparable firms that had not. They were not cutting people. They were hiring more of them, and faster.

Source: CSIRO / Australian Journal of Labour Economics, April 2026

The Irony at the Heart of the Jobs Debate

The finding that will produce the most discomfort in certain quarters concerns what the CSIRO researchers call "AI-exposed" roles — occupations that include tasks AI systems can now perform. These are not the manual jobs that previous waves of automation displaced. They are professional and knowledge-intensive roles: accountants, lawyers, analysts, the kind of people who spent years acquiring expertise and reasonably expected that expertise to remain valuable.

The study found that demand for these roles did not fall in firms that had adopted AI. It fell — slightly, but statistically significantly — in firms that had not adopted AI. Which is, if you follow the logic carefully, a rather extraordinary finding. The people most at risk from AI-driven displacement are not those working inside organisations that have embraced the technology. They are those working inside organisations that have not.

— Dr Claire Mason, lead researcher, CSIRO Workforce & Productivity Research Team

This is the kind of counterintuitive result that rewards a moment's reflection. An accountant working at a firm that uses AI to handle data processing can spend their time on judgement, analysis, and client relationships. An accountant working at a firm that has decided AI is too complicated or too uncertain will, in due course, find themselves competing against the first accountant — and losing.

The Deskilling Myth, Firmly Retired

Another piece of received wisdom that does not survive contact with this data is the deskilling hypothesis: the idea that AI, by absorbing the complex parts of jobs, will gradually reduce the need for human expertise and leave workers doing simpler, lower-value work. It is a coherent theory. It is not, on present evidence, what is happening.

Across the dataset, job advertisements began listing more skills over time, with the increase most pronounced in AI-adopting firms and in AI-exposed roles. The jobs are not becoming simpler. They are becoming more demanding — requiring workers to bring greater capability to the table, including, increasingly, the ability to work effectively with AI tools.

Dr Mason's observation about where AI skills are now appearing is worth pausing on. Her research found AI-related requirements emerging in job postings for sales representatives, security officers, and architects. These are not roles that feature prominently in the standard AI-disruption narrative — which has always been more comfortable talking about programmers and data scientists. The implication is that the question of who needs to understand AI is expanding rather rapidly, and the answer is converging on "basically everyone."

The Real Competitive Divide

What the CSIRO research describes, with the calm authority of actual data rather than speculative modelling, is a growing schism in Australian business. Not between humans and machines, as the popular framing would have it, but between firms that have made the decision to engage with AI and those that are still deciding. The former group is hiring more, asking more of their people, and building organisations that are structurally more competitive. The latter group is, slowly and without necessarily noticing, falling behind.

This divide is not yet dramatic. The data does not show AI-adopters wiping the floor with everyone else in a single dramatic quarter. What it shows is something more insidious and ultimately more consequential: a consistent, compounding advantage that accumulates quietly over years. The tortoise is already moving. The hare is still explaining why running is probably overhyped.

On Being a Domain Expert in the Age of Assistance

Dr Mason's research also contains a finding that ought to be genuinely reassuring for anyone who has spent years developing deep expertise in a field — and has been wondering whether that expertise is now redundant. The latest AI tools are, she observes, very user-friendly. But using them well takes knowledge, judgement, and experience. Domain experts — those with hands-on expertise in a field — tend to get the strongest results from AI.

This is not the "AI will handle the boring bits and humans will do the creative bits" platitude that has been doing the rounds since the technology first became visible. It is a more specific and more interesting claim: that the value of deep expertise is not diminished by AI but amplified by it. The person who understands their field well enough to direct an AI effectively, evaluate its outputs critically, and apply its results with good judgement is more capable, not less, than the same person without the tool.

Source: CSIRO / Australian Journal of Labour Economics, Dr Claire Mason, April 2026

What This Means for the Business Owner Reading This

The CSIRO research covers the period from 2020 to 2023 — before the current generation of generative AI tools became widely used. Dr Mason notes that more recent labour market data shows similar patterns continuing. The findings are, if anything, likely to become more pronounced as the tools become more capable and more embedded in everyday work.

For the small business owner or manager who has been watching the headlines and concluding that AI is either a job-destroying catastrophe or an overhyped distraction, the research offers a more useful frame: it is a capability investment with compounding returns. The firms that made it early are already ahead. The gap will widen.

The question is not whether your industry will be affected. The CSIRO data shows AI-related skill requirements appearing in sales, security, architecture, and trades — the "AI is only for tech companies" argument has not survived contact with reality any better than the "AI will eliminate all the jobs" argument. The question is whether the capability being built inside your business is keeping pace with what is being built inside your competitors'.

The Reasonable Conclusion

The CSIRO study is careful to note that it does not claim AI is risk-free, or that disruption will not occur. Jobs are changing. Tasks are being redistributed. Some roles will evolve in ways that not everyone will find comfortable. These are real observations, and they deserve honest acknowledgement rather than dismissal.

But the data does not support the most common version of the AI anxiety story — the one in which the machines arrive, the humans depart, and the economy contracts around a shrinking pool of work. What it supports is a considerably less dramatic, and considerably more hopeful, story: AI-adopting firms need more people, not fewer. Those people are asked to do more, and equipped to do it. The work is becoming richer, not simpler.

The threat to jobs in the AI era, on the evidence currently available, is not that your employer will replace you with a machine. It is that your employer will fail to use the machine — and then discover, too late, that the firm down the road has been using it rather well for several years.

Australia's national science agency has, with admirable rigour, told us which side of that divide is better to be on. The sensible response is to act accordingly.