We have become accustomed to talking about artificial intelligence as though it were weather.

AI is coming. AI is transforming everything. AI will change our jobs, our businesses and our economy.

The language sounds descriptive, but it does something more subtle – it makes the future feel inevitable. And when something feels inevitable, we stop asking who is making the decisions that determine what happens next.

The technology itself may well be inevitable, its consequences are not. Yet the language of inevitability quietly collapses the two, making acceptance of the first feel like acceptance of the second.

That is what makes the Albanese Government’s AI announcement last week so interesting.

In unveiling its plan for AI, including a new Office for AI and a national standard, the Prime Minister made a case that went beyond simply regulating the technology.

He argued that Australia could not “turn back the clock or press pause” on AI, but that the country could still decide how it would be developed and deployed here.

Albanese argued that Australia can “set the terms” and “determine AI’s social licence”, but warned that the country must act now before companies have “built whatever they want, wherever they want” and Australia is left trying to reopen negotiations.

That is an important shift in the conversation. It moves us beyond the question of how Australia regulates AI once it is here toward the more fundamental question of whether Australia has agency over what AI becomes here.

The most important decisions about AI are not being made somewhere in a distant future. They are already happening, often quietly, before the rules intended to govern them have even been written.

Consider the extraordinary scale of investment now flowing into AI infrastructure.

Back in March, $51.9 billion in data-centre projects were fast-tracked through NSW planning processes. Binding rules governing data centres and their impacts are not expected until 2027.

The point is not whether those projects should or should not proceed, it is that the order of events matters. The decisions are being made first, the framework comes later.

These frameworks involve different governments and different processes, which is rather the point. The decisions shaping AI are distributed, fast-moving and already underway, while the framework intended to govern them is singular, federal and still two years away.

This is precisely the risk Albanese himself identified. His warning about not waiting until companies have built “whatever they want, wherever they want” is not just a question of regulation. It is a question of timing – when do we exercise our agency, and when do we discover that the most consequential decisions have already been made?

Once the rules are announced, the announcement itself can become the story. The public sees a government responding. Headlines appear. Policy is debated. The issue feels as though it has been dealt with. But the underlying decisions may already have happened.

I’ve spent fifteen years in behavioural research, and this pattern has a simple explanation. The promise of a rule and the existence of a rule are different things, but we don’t store them differently.

Both land in the same mental bucket, and the bucket reads ‘solved’.

Writing it on the to-do list feels like progress. Buying the running shoes feels like the training. Announcing the rule feels like the rule itself.

This is how agency disappears from public view. Not because no one is making decisions, but because the decisions are made before most people recognise that there was a decision to make.

The same dynamic is playing out in the way we talk about AI.

When we describe AI as an unstoppable force, we obscure the choices that shape it. We forget that AI systems are designed, funded, trained, deployed and regulated by people and institutions.

We forget that infrastructure has to be approved, energy has to be allocated, data has to be sourced and workers have to be retrained or displaced.

We also risk a form of AI washing – presenting AI as an inevitable force of nature rather than a technology shaped by human choices, commercial interests and political decisions.

The Prime Minister is right, this is our time to decide what AI looks like in Australia. But that decision cannot be reduced to the rules that govern AI systems once they are deployed.

It also has to encompass the choices being made now about the infrastructure that will power them, the resources they will consume, the data they will rely on and the economic value they will create.

That is where the conversation about AI agency becomes more important than the conversation about AI regulation alone.

The European Union and the United Arab Emirates offer two very different examples of what it means to make that agency explicit.

The EU has pursued technological sovereignty through regulation, infrastructure and efforts to reduce its dependence on foreign technology.

The UAE, by contrast, has taken a more state-led approach, investing heavily in AI infrastructure, talent and adoption to position itself as a producer and shaper of the AI economy rather than simply a consumer of technology developed elsewhere.

Their political systems and strategies are very different, but underneath them is the same fundamental question – who gets to decide the terms on which AI enters a society and an economy?

Australia needs to ask itself that question too.

Not because we can stop AI – we cannot. Not because every decision about AI should be made by government – it shouldn’t.

The question is whether Australia will be an active participant in shaping what comes next, or whether we will allow the most consequential choices to be made by default, by whoever moves first, whoever controls the infrastructure or whoever has the capital to build at scale.

The Albanese Government’s announcement is therefore an important starting point, but the test of agency is not simply whether Australia can write the rules. It is whether we are making the decisions early enough to shape the future those rules are intended to govern.

The real challenge, then, is not convincing people that AI is inevitable. It is convincing them that they still have agency over what comes next.

And the first step is simple. Stop watching the speeches, start watching the queue.

The deciding has already started and it will not wait for 2027.