The Australian conversation about AI has settled around risk and regulation. Its economic promise will turn less on the rules Australia writes than on the capability built around the technology.
Much of Australia’s discussion about artificial intelligence now runs through the compliance layer. Guardrails, risk registers, safety obligations and the shape of a possible AI Act take up most of the attention. These are reasonable concerns, though together they add up to a defensive stance, one preoccupied with containing a technology rather than learning to use it well.
However, the official signals are starting to move. In its 2025 review of data and digital technology, the Productivity Commission argued that economy-wide AI rules should be a last resort. It cautioned that poorly judged regulation could stifle a material productivity gain. The National AI Plan of December 2025 shifted the emphasis from mandatory guardrails to adoption.
Compliance guards against harm. What it cannot do, on its own, is generate value. The return on AI depends less on the model than on the setting it enters: the skills, the data, the management practices and the organisational change that turn a capability into a result. That dependence sits at the core of the AI Strategic Complementarity Principle set out in the work of the Acton Institute for Policy Research and Innovation.
The Productivity Commission reaches a similar conclusion by a more economic route. The complementary investment needed to make AI pay can run to several times the cost of the technology, and the payoff often lands years later. The constraint is seldom the capability itself. More often it is the surrounding conditions that lag.
The past year has made the point in public. Agent deployments have stalled inside organisations that lacked the data, the processes and the oversight to hold them up, and several courts have penalised submissions resting on AI-fabricated citations. The models did what they were built to do. What gave way was the verification, the judgement and the data discipline that ought to have surrounded them.
For Australia, the gains are more likely to come from augmentation than from wholesale automation. Augmentation lifts the productivity of human judgement instead of dispensing with it, which suits an economy built on services and smaller firms. The Productivity Commission puts the possible uplift at around 4.3 per cent in labour productivity over a decade, with services among the main beneficiaries.
There is a genuine base to work from. The government’s own stocktake identifies capability in computer vision, multimodal systems, smart sensors and field robotics, alongside high value uses in health, agriculture and advanced manufacturing. Industry puts the prize higher again, though those numbers come from parties with an interest in the answer and might be read with some caution.
The promise is real; the supporting capability is thin. Three weaknesses stand out, and none is new to anyone who has followed Australian innovation performance over the years. They are worth stating directly, because a compliance-first habit can deepen them, treating AI as a hazard to be managed rather than a capability to be grown.
- Business research and absorptive capacity. Business spending on research and development has sat at roughly 0.9 per cent of GDP since 2017–18, about half the OECD average. Firms that spend little on knowledge tend to be slow to take up what is new.
- Management capability. The gap between the best run firms and the rest in the everyday practices that make technology pay, what the Acton Institute for Innovation calls the management chasm, may count for more with AI than access to the models.
- Data and diffusion. Access to good data is patchy, and the benefits of AI usually reach small and medium firms last. The national result will be decided more by how widely these tools spread than by who first invents them.
A move from compliance to opportunity does not require dropping any safeguard. It asks that the conditions for adoption be given at least equal billing. Policy can act on the complements head-on: skills and management capability, trusted data infrastructure and help for smaller firms to take the technology up. The AI Adopt Program and a larger National AI Centre make a start.
Government can lead by putting AI to work in its own operations. The Australian Public Service AI plan and the GovAI platform cast the public sector as an early user of AI rather than only its regulator. By modelling what careful practice looks like – what departments buy and how they use it – may shape adoption across the economy as much as any statute.
These moves work better together than apart. Skills, data, diffusion, research incentives and public-sector use feed one another, and their value builds when they are timed and sequenced as an innovation ecosystem. Run as a single program rather than a run of announcements, they begin to add up to a national capability rather than a set of gestures.
Safety and growth are not the real alternatives. The choice is between a stance that contains AI and one that builds the capacity to use it well. On the second path, AI raises the output of clinicians, farmers, engineers, teachers and public servants, with human judgement still doing the deciding. Whether Australia gets there rests on the supporting conditions, and those are a matter of deliberate policy.
Compliance will stay part of the job. On its own it will not produce the gains the current debate keeps describing. The more practical question is what Australia should build around AI so that its benefits are broadly shared. That is where policy attention could now usefully turn.
John H Howard
John H Howard is a researcher, policy analyst, management adviser, and author with three decades of experience advising governments, universities, and industry on science, research, and innovation policy and strategy. He is Executive Director of the Acton Institute for Policy Research and Innovation and an Honorary Visiting Professor at the University of Technology Sydney. He can be contacted at john@actoninstiyute.au