Australia’s central bank published the receipt today, and it is not the one the AI industry has been handing out. Sarah Hunter, the Reserve Bank’s Assistant Governor (Economic), told a Sydney conference that there is only limited evidence the technology has materially lifted productivity in this country. On the same morning, her colleagues released the bank’s first official monitor of where AI is already changing the labour market — and the answer is job adverts, in specific occupations, for specific people.
That is the trade Australian workers are being asked to accept. Take the disruption now — the smaller hiring round, the graduate with no first rung — and wait for the productivity dividend that was supposed to pay for it. The dividend has not shown up. The disruption has.
Sarah Hunter — The Payoff Is Missing, and She Said It in Public
The speech, delivered at the UBS Australasia Conference, is the most useful document the bank has produced on this subject because it refuses to sell anything. Hunter laid out adoption first: the share of firms reporting no AI use at all has collapsed from roughly a third in 2025 to almost nothing in 2026, while limited adoption climbed from about 40 per cent to roughly 60 per cent and deep, transformative use sits near 10 per cent. Nearly everyone has bought in.
Then the other column. Measured productivity gains remain thin, and Hunter was explicit that the aggregate numbers do not yet show the payoff. She reached for the standard explanation — the productivity J-curve, the lag every general-purpose technology produces while firms rebuild their processes around it, the way electricity only paid off once factories were redesigned rather than merely re-powered. It is a fair analogy. It is also an admission that the bill is arriving years before the benefit, and that someone has to hold it in the meantime.
The range of expert estimates she cited makes the waiting worse, not better: forecasts for AI’s contribution to annual productivity growth over the next decade run from more than 2 per cent down to 0.1 per cent. That is not a forecast. That is a shrug with footnotes.
35 Per Cent — The Firms That Bought the Future and Got Nothing Back
The bank’s own liaison data is blunter than any think-tank projection. Around 35 per cent of firms report no productivity gain at all from AI. Another 30 per cent report gains they describe as minor. Roughly 23 per cent call theirs moderate, around 11 per cent significant, and a sliver say they have gone backwards.
Read that as a business case and it fails. More than a third of Australian firms are paying for a capability that has returned nothing measurable, and only about one in nine can point to a gain worth the name. The sales deck promised transformation. The ledger says most buyers got a slightly faster inbox.
Five Occupations — Where the Adverts Actually Dried Up
The research note published alongside the speech, by RBA staffer Mike Major, is the part working people should read. It sorts occupations by how exposed their tasks are to automation, benchmarks everything to November 2022 — the month ChatGPT put generative AI in front of the public — and then watches what actually moved. What moved first was hiring.
New job advertisements have fallen across the board since that benchmark, but they fell furthest in the most AI-exposed group. The occupations driving the decline in Commonwealth job ads are specific and unglamorous: general clerks, software and applications programmers, ICT business and systems analysts, call and contact centre workers, and receptionists. Employment growth and total hours worked have also run a little weaker in the more exposed occupations than the less exposed ones.
The note is honest about its own limits, which is what makes it credible. It finds no clear divergence in job-finding rates or in the rate at which people move from employment to unemployment. Its complete occupation data only runs to February 2026. And exposure is not the same as displacement — a task that AI can perform is also, often, a task AI can help a person perform faster.
Still, the direction is consistent, and it points at the doorway rather than the exit.
The Graduate — Paying Full Price for a Bet Somebody Else Placed
Here is the human cost, and it is not a layoff notice. It is a 22-year-old who did the degree, built the portfolio, applied to forty junior roles and got silence. The RBA note flags softer outcomes for young people and graduates, and notes that international evidence has tended to find weaker hiring and job-finding for entry-level workers in exposed occupations rather than mass redundancies.
That distinction matters enormously to the person living it. A layoff is visible, counted, and often cushioned. A shrunken graduate intake is invisible — no headline, no severance, no alumni fund. It simply becomes a year of a young person’s life that does not start. Entry-level work is the apprenticeship ladder of every profession, and the monitor says the bottom rungs are the ones AI exposure touches first.
10-Fold — The Build-Out Arrives Years Before the Bill
While the productivity gain lags, the spending does not. The bank estimates realised data centre investment has risen tenfold over three years, admittedly from a very low base, and much of that money leaves the country as imported equipment, which trims the net contribution to national output. Money is being spent here. Value is being booked somewhere else.
Hunter also put a second risk on the record that deserves more attention than it got: if the returns on the global AI investment boom disappoint, asset prices — equities and corporate debt among them — are exposed to sharp repricing, at a moment when risk premiums are historically thin. In plain terms, the same bet that has not yet paid workers has also been priced into portfolios. The technology can underdeliver on jobs and still damage the people who own the shares.
None of this is an argument that AI is a fraud. Adoption at this speed is real, and the productivity J-curve is a genuine historical pattern. It is an argument about who carries the risk during the lag — and the RBA’s own numbers say it is the clerk, the receptionist, the contact-centre worker and the graduate, while the firms booking the efficiency gains are still waiting to find them.
The bank built a canary. The canary is alive. It has also stopped singing in five occupations, and nobody in the sales deck mentioned that the cage would be paid for by the people inside it.
Sources: Reserve Bank of Australia, RBA Insight 2026-09, Reuters, RBA Q&A transcript.