What the RBA said
On 9 November 2026, Sarah Hunter, the Reserve Bank of Australia's Assistant Governor (Economic), gave a speech at the UBS Australasia Conference in Sydney titled "What Happens to the Economy as 'Intelligence' Gets Cheaper?". Her main finding is that there's limited evidence AI has materially lifted productivity so far, and limited evidence of widespread job losses (Reuters via The Star; InnovationAus). The speech treats AI as a general-purpose technology, the same category as electricity and the internet, whose economic effects arrived slowly and were hard to forecast at the time (RBA).
For people building AI products, the data behind the speech is more useful than the headline. Much of it comes from the RBA's liaison programme, the bank's regular interviews with Australian businesses. Those interviews show how far Australian firms have actually got with AI. Australia is a market many New Zealand startups sell into, so the figures describe a lot of their buyers.
Adoption is broad and shallow
The share of liaison firms reporting no AI adoption fell from about a third in 2025 to almost zero in 2026 (RBA). Most of that movement went into the lowest category. Firms reporting limited adoption rose from about 40 per cent to about 60 per cent, and moderate adoption rose from about 20 to about 30 per cent. Deep or transformative adoption only moved from about 8 to 10 per cent (RBA). The speech doesn't define these categories or say how many firms took part, so we'd read the figures as showing which way things are moving, with soft edges on each share.
Reported productivity gains follow the same pattern. About 35 per cent of firms report no gain, about 30 per cent report minor gains, about 23 per cent moderate gains and about 11 per cent significant gains. Very few report losses (RBA). Hunter's examples will be familiar to anyone shipping AI tools. Lawyers can review and summarise large sets of documents faster. Software firms use AI agents to produce batches of code, and their engineers spend more time overseeing that output (RBA).
Taken together, the two charts point to a large group of Australian firms that have bought or trialled AI tools without changing their workflows enough to see a measurable return. We'd expect that gap to shape how these firms buy. A firm in the limited group can already say it uses AI, so the next AI product it buys has to show a productivity gain the firm can measure.
The software engineering estimate
The speech's strongest productivity figure comes from the United States. Hunter cited a study of share prices which suggests markets expect AI to permanently raise software engineering productivity by around 30 per cent (RBA). The study is NBER Working Paper 35793 by Blumenfeld, Hazell, Lian and Schaab. It measures how sensitive each firm's share price is to an AI stock index, and how that sensitivity changes with the share of the firm's payroll spent on software engineers. For November 2022 to December 2025, the authors put the effect at the equivalent of a permanent 32.6 per cent rise in software engineering productivity. In their baseline case, that would raise the level of US GDP by 3.6 per cent. They report that by mid-2026 the estimated effect had more than doubled from its end-2025 level, driven by progress in coding agents (NBER).
Builders should keep the method's limits in mind. The paper measures what investors expect, so its figure follows market sentiment about coding agents and will move when that sentiment moves. The speech also quotes estimates of AI's effect on annual productivity growth over the next decade that range from more than 2 per cent down to 0.1 per cent (RBA). A range that wide shows how little agreement economists have reached.
Jobs: small effects, mostly in exposed roles
About three quarters of liaison firms report no change in headcount from AI. Around 12 per cent report slower hiring or roles left unfilled, about 8 per cent report lower headcount and about 6 per cent report higher headcount (RBA). Hunter described "tentative signs of adjustment": weaker job ads and employment in occupations most exposed to AI, and stronger growth in less exposed areas, especially for younger workers (Reuters via The Star). Since late 2022, employment of 15 to 24 year olds in the most exposed occupations has grown slightly slower than employment of older workers. The speech also cites Jobs and Skills Australia figures. They put about 4 per cent of workers as highly exposed to AI doing their work for them (automation) and about 31 per cent as highly exposed to AI helping them do it (augmentation) (RBA).
Data centres, inflation and interest rates
Data centre investment actually completed in Australia has risen about tenfold over three years, from a low base. Much of that money goes on imported equipment, which shrinks what it adds to GDP (RBA). Hunter argued that AI spending is adding to demand now, while the extra supply it creates arrives later. The short-term effect is therefore inflationary, and she expects it to fade as capacity and productivity catch up (RBA). Reuters reported that the RBA has raised interest rates five times this year, taking the cash rate to 4.85 per cent, an 18-year high. Reuters named the data centre boom as one reason for the increases (Reuters via The Star).
On consumer prices, the speech found the increases are concentrated in a narrow group of electronics such as personal computers. Capacity pressure in the semiconductor supply chain is passing through to some of those goods (RBA). The speech also flagged a risk to financial stability. Investors are currently demanding less extra return for risk than usual, and if returns on AI investment disappoint, share prices and corporate debt could fall sharply (RBA).
What the data means for New Zealand builders
The speech doesn't mention New Zealand, and we aren't aware of comparable published New Zealand survey figures, so applying the numbers here is our own inference. Australia is the nearest export market for most Kiwi software firms. The liaison data gives a reasonable picture of a typical Australian buyer in late 2026: already using AI in light ways, and mostly not yet seeing much return.
That picture has practical consequences for product decisions:
- Products pitched at a firm's first use of AI are chasing a group the RBA now puts at almost zero.
- Products that move a customer from limited to moderate adoption need evidence of measured gains. Building usage and outcome metrics into the product from the start gives sales teams firm numbers to show.
- Teams whose costs depend on compute should watch the RBA's inflation point. The bank sees semiconductor capacity pushing up some hardware prices, and those pressures can reach server and GPU costs.
- With Australian rates at an 18-year high, buyers there pay more to borrow. We'd expect them to look harder at how quickly an AI project pays for itself.
The RBA says understanding AI's effects is an ongoing research priority. The speech points to the bank's research agenda and to a set of RBA Insights papers published in November 2026 on AI and the labour market, export prices, data centre investment and household wealth. Those papers, and next year's liaison figures, will show whether firms in the limited-adoption group start moving up. If your own customer data shows that shift, it's worth bringing to an ASAI event, where other builders are working through the same question.
Sources
- What Happens to the Economy as 'Intelligence' Gets Cheaper? (Reserve Bank of Australia, speech by Sarah Hunter, 9 November 2026)
- RBA's Hunter says AI has yet to deliver productivity boost, trigger job losses (Reuters via The Star)
- AI yet to deliver productivity boost, trigger job losses: RBA (InnovationAus)
- The Macroeconomic Effect of AI: Sizing the Software Engineering Channel (NBER Working Paper 35793)
Build with people working on the same things
Join the Auckland Society for AI: NZ$599 a year with a ForgeBox and the members' projects area. Our Auckland meetups are open to everyone.