Something interesting
is happening in
Australian AI

Most of the world is watching OpenAI, Anthropic and Google take turns releasing the most powerful model ever built, roughly every six weeks. Meanwhile some of the more interesting Australian companies have quietly gone off and done something else. They’re using AI on problems in industries we already understand. Mining. Energy. Engineering. Healthcare. Robotics.

I suspect this is a much bigger opportunity for Australia than trying to build our own ChatGPT. I might be wrong about that, and I’ll come back to why.

The race we probably don’t need to win

The largest technology companies on Earth are spending extraordinary amounts on data centres, chips and ever larger models. The figures have stopped behaving like money and started behaving like weather.

Australia is in the race too. Billions have been committed to AI infrastructure, and there is growing interest in us as a place to put very large computers, presumably because we have a lot of room and the computers don’t mind the heat.

But I keep coming back to one question. Do we need to own the technology underneath AI, or should we concentrate on what we can actually do with it?

Now, the people building the infrastructure would say, quite reasonably, that you can’t do much with technology you don’t control, and that a country which rents all its computing from three American companies has made a decision about its future whether it meant to or not. I think that’s right, as far as it goes. There are good reasons to invest in local computing and sovereign capability and I’m not arguing against any of them. What I’m arguing is narrower. Australia is not going to outspend the American tech giants on general-purpose foundation models. We can be sovereign about where the computers sit without pretending we’ll win the race to build what runs on them. The sooner we’re comfortable saying that out loud, the sooner we can get on with the interesting bit.

The companies that caught my eye

Adelaide-based Metacognition AI has expanded its pre-seed round to A$15 million. It’s not building another conversational assistant. It’s working on AI that can reason, remember and operate in the physical world, which may eventually help a robot understand an instruction and then go and do something about it.

Closer to home, WA’s Augment Technologies was recognised at the 2026 Innovators of the Year awards for technology that combines physical modelling with AI to track where valuable ore ends up after a blast.

Tracking rocks doesn’t sound as exciting as a robot writing poetry. But ore, it turns out, does not stay where it is put. It has been wandering off after blasts for the entire history of mining, and only now is anyone in a position to ask it where it went. If a mining company knows more precisely where its ore went, it recovers more of it and wastes less. That has an obvious dollar value, and it’s a problem that needs geology, engineering and an understanding of how a mine actually runs on a Tuesday. You cannot solve it by wiring a language model to a text box and hoping.

What experience brings

I’ve spent most of my career developing technology and working with engineering teams, a good part of it in medical devices.

A few years ago I was part of a team getting ready to start a clinical trial on a Monday. The device was a new insulin pump. The participants were people living with diabetes who had signed up, been through the consenting, and were, by the Friday, properly excited about trying it. On the Friday afternoon we stopped the trial before it started. The pump was alarming too often. Not dangerously, just constantly, in the way a smoke detector with a tired battery is not dangerous and is also not something you can live with. The pump worked. It had worked in every demonstration anyone had given. It just hadn’t yet met a weekend.

That’s the lesson the medical device world teaches you, usually at about four o’clock on a Friday. Getting something to work in a demonstration is the beginning. Getting it to work reliably, in someone’s actual life, is a different job. A device doesn’t become useful because the technology inside it is clever. It has to solve a real problem, people have to be able to use it, someone has to be able to make it, and you need evidence that it does what you say, because at some point a person will arrive with a folder and the folder will have questions.

AI is running into the same realities now. It has become remarkably easy to build something impressive. Getting it into a workplace, alongside the systems already there, and earning enough trust that someone will let it near a decision that matters is still hard. That’s where domain experience counts, and Australia has a lot of it.

The advantage we may be overlooking

Think about the industries Australia is actually good at. Mining and mineral processing. Remote operations and autonomous equipment. Agriculture. Energy. Healthcare and medical technology. Advanced engineering.

These industries have expensive problems in difficult places. They also generate enormous amounts of data, most of which has been sitting on a server somewhere, patiently, waiting for someone to have the time. AI gives us new ways to read that data, find the patterns and automate the tedious parts. But the AI is only part of it. Understanding the problem is the competitive advantage.

Someone who has spent twenty years running a mine understands an expensive problem a brilliant AI engineer would never think to look for. A clinician sees an opportunity that generic healthcare software will sail straight past. An experienced manufacturing engineer knows which step in the process has been quietly eating the margin since 2014, and has probably mentioned it in three meetings nobody wrote up.

Combine that knowledge with increasingly accessible AI and something happens. People who have spent decades building expertise in established industries suddenly have a new set of tools, and nobody has told them they aren’t allowed to use them.

I’m testing some of this myself

Part of my interest is that I’m running the experiment on myself. Through projects including Quokkacorn and Lensmilk I’m finding out how much one person, or a very small team, can build with AI. Lensmilk is an attempt to rethink advertising production, from the original brief through concepts, storyboards and video. It’s a long way from regulated engineering, which is part of the appeal. Nobody in advertising has ever asked me for a risk management file.

I should report how it’s going. The first piece of user-generated-style content Lensmilk produced was supposed to look like a natural, relaxed influencer talking about a product. It looked like a nervous robot that had been told about influencers and was doing its best. The product was fine. The robot was not fine.

So AI lets me try ideas that used to need a much bigger team and considerably more money, and I’m learning that being able to build something is not the same as having a business. You still need to understand the customer and solve a problem they’ll pay for. You still need to deliver reliably and to reach people, who are busy and have not been waiting for you. AI makes some of that easier. It doesn’t make any of it optional. If anything, as building gets cheaper, understanding customers and their problems becomes the part that’s actually hard.

An opportunity for people who’ve been around a while

There’s another angle that doesn’t get enough airtime. Most of the AI conversation is about young founders, new startups and people learning to code. What about people who have spent twenty or thirty years in an industry?

They’ve accumulated experience, relationships and a detailed understanding of problems that are invisible to outsiders. Historically, turning that into a technology business meant raising substantial funding, hiring developers and spending years building a product, by which point the problem had usually moved. The cost of simply testing an idea is now falling fast. That doesn’t guarantee success, but it does change who gets to have a go.

Perhaps the more interesting consequence of AI won’t be a generation of twenty-year-old entrepreneurs. Perhaps it will also be a generation of experienced people finally able to solve problems they’ve understood for decades, not necessarily chasing the next billion-dollar company, just building something useful and profitable that gives them a bit more independence. I find that far more compelling than another chatbot.

So what should Australia actually do?

We should absolutely support fundamental research, advanced computing and companies developing new AI technology. But we should also recognise the opportunity that is sitting right in front of us, and has been for some time, with its arms folded.

Australia has an enormous amount of expertise in industries where the problems are real, expensive and difficult, and we now have access to technology that dramatically lowers the cost of trying solutions. The opportunity is to put those two things in the same room.

Not every successful Australian AI company needs its own foundation model. Some might recover more ore from a mine. Others might cut wasted electricity, give clinicians more time with patients, or make industrial equipment safer. Some will be built by surprisingly small teams. Some will stop a trial on a Friday because the alarms wouldn’t shut up, and start again a month later with a product that works.

That’s the future I’m interested in. Not what AI can do, but what we can actually make better with it.