Treasury to Chalmers: Australian AI is widespread and shallow

Two-thirds of Australian businesses have touched the tool. Fewer than one in ten have done anything that counts. You've probably briefed "we have AI" on the strength of a licence and a pilot that never left the team that bought it. That is not a skills shortage story yet. That is a pretend-adoption story.

Treasury told Treasurer Jim Chalmers the country has widespread and shallow AI. Not a vendor survey. Around two-thirds of Australian businesses report some AI adoption. Less than 10 per cent report significant adoption. That is the country in one pair of numbers.

Treasury's point is that you do not get the productivity if you do not rebuild organisational capital: processes, business models, management practices and workforce skills. The jobs wipe-out has not shown up in the advice. Chalmers called it the biggest economic transformation in our lifetime and said we cannot sit around hoping the benefits fall into our lap. The useful instruction is that the country cannot wait for the benefits to arrive on their own. Process, management practice and workforce skills are what move the adoption number.

The people who said the quiet part

Brian Collins at Enterprize Tasmania said it is more a confidence gap than a skills gap. Warren Hogan at Judo Bank said the risk is displacement without a productivity lift. Those two sentences should be taped to every CAIO's monitor. You can train people into a hole if the process underneath is still the 2019 one.

Treasury's long-run growth assumption in the piece is 1.2%. Upside productivity 1.5 to 2%. Downside 0.5 to 0.8%. Data centres: about A$150 billion by 2030, 162 operational as of March, 130 proposed, about two-thirds of the spend on imports. Defence Minister Richard Marles is heading to San Francisco to see Anthropic, OpenAI, Microsoft, Google, Amazon and NVIDIA. Andrew Charlton is at the G20 on AI skills. Ministerial travel and data-centre spend do not fix the less-than-10% significant-adoption problem.

If you are an Australian CHRO, this is your brief from the centre. Some-AI is the two-thirds. Significant is the tenth. The gap is process and skills, on Treasury's own list. A named Australian version of sitting in that work is the Australian Unity engagement. Run it as process redesign, not as a module that lets you tick we-trained-them while Hogan's displacement-without-lift scenario quietly happens in the contractor line.

Collins's confidence gap is the L&D trap. People will not use the tool in the live process if they think they will be punished for being wrong. A course does not fix that. A manager who will sit with them on the hated workflow might. Hogan is describing the failure mode if you automate the person and leave the process: fewer people, same output, worse.

The 1.2% long-run assumption versus 1.5 to 2% upside and 0.5 to 0.8% downside is why Chalmers cannot sit around. Those are growth rates, not training KPIs. Still: Treasury listed workforce skills next to processes and management practices. If your AI paper to the executive team has a model diagram and no skills line, you are not reading the same advice the Treasurer got.

Data-centre billions and Marles in San Francisco will get the cameras. They do not move the less-than-10% significant-adoption number. Charlton talking AI skills at the G20 is closer. Copy that, not the campus render.

The jobs wipe-out "has not materialised" line is easy to over-read. Treasury is not promising your contractor bench is safe. It is saying the apocalypse slide is not in the advice. Hogan's displacement-without-lift is the quieter risk and the one a CHRO can actually work. Significant adoption under 10% means most firms have not even earned the right to that conversation yet. They have a licence and a story.

What this means for L&D

Two numbers. One brief. Two-thirds / under 10% as the opening slide. Name Chalmers, Hogan and Collins. Put workforce skills in the same sentence as processes. Do not brief the data-centre billions as if they were a training budget. Widespread is not significant. The programme that counts is the one that rebuilds the process, not the one that ticks we-trained-them.

Sources: ABC News, 31 August 2026.