NOTE: This speech was delivered on Thursday 22 October 2026. Check against delivery.

Opening

Good morning, and thank you for having me.

It’s a pleasure to be in a room full of people who chose public service because they wanted to make a difference.

That purpose hasn’t changed. 

But we’re being asked to deliver with tighter resources and rising expectations. In that context, innovation can feel like a luxury.

I see innovation as a necessity, not a luxury. It’s about finding better ways to deliver what people need, not chasing novelty.

Today, I’d like to begin with one thought. I hope you’ll carry it with you for the next twenty minutes – and hopefully longer!

Technology is making is easier to move fast.

But direction matters more than acceleration.

I’ve spent much of my career around large institutions and complex reform. I’ve seen government at its very best: people coming together from across government, putting aside who’s responsible for what and solving a problem because the community needed them to. 

I’ve also seen good ideas lose momentum — not because they were bad ideas, and not because people lacked commitment, but because the path from idea to implementation was simply too hard.

Here’s what I’ve come to believe about innovation in government. It isn’t really about having more ideas. Public servants are full of ideas.

It’s about whether our systems can recognise a good idea, test it safely, scale what works, and stop what doesn’t — before the need has moved on.

That’s where the APS AI Plan comes in. It gives us a shared direction, built around three pillars: trust, people and tools. Those foundations matter because having the technology is only part of the job. 

We need people who can use it confidently, safeguards we can trust, and a clear sense of which problems we’re trying to solve. The challenge is to bring those things together so that a good idea can become part of how we work, not just another pilot.

So what does that look like in practice? I’ll start with what we’ve already built and what still gets in the way. Then I’ll talk about four practical choices: choosing a problem worth solving, making the smallest useful move, reusing what works and making it safe to go first. 

I’ll finish with a small ask of each of you.

Where we’ve got to

So, let me start with what we’ve already built.

Think about an ordinary Tuesday. A pension is paid. A passport is processed. A weather warning reaches a community before the storm does. Someone calls for help, and the system answers.

None of that is visible when it works. And that’s the point. 

Nobody throws a parade for the payment that arrived on time, or the cyber incident that was contained, or the system that didn’t fall over.

But it’s real. People in rooms like this one built it. And it’s a tradition of adaptation. We’ve moved from paper to digital, from counters to platforms and, at our best, from designing services for people to designing them with people. 

The OECD’s Digital Government Outlook 2025 recognised the technical maturity of Australia’s digital reforms.

So the question isn’t whether we can adapt. We’ve shown that we can. The question now is how we use the possibilities technology gives us to focus on the challenges that matter most.

What gets in the way — and what we can change

So what gets in the way — and what can we change? 

Because we’ve achieved so much, we owe it to each other to be honest about what slows us down — and equally clear about what is within our power to change.

Many of our rules, processes and funding models were built for a world that moved more slowly. 

They were sensible answers to the questions of their time. Annual funding cycles make sense when needs change annually. Long approval chains make sense when information travels on paper. Agency-specific systems make sense when problems arrive with one department’s name on them.

But a family getting back on its feet doesn’t experience housing, health, employment and social security as separate portfolios. A small business doesn’t care which agency owns which bit of a process. And technology doesn’t wait for the next budget cycle.

You’ll recognise how that feels from the inside. A promising idea that takes longer to clear in the system than it took to develop. 

A pilot that delivers something useful, then stalls because nobody funded the unglamorous work of integration, support and ownership. 

Or the workaround that started as a clever fix and, 5 years later, has become the system.

And standing still isn’t the low-risk option. Ageing technology gets more expensive and more fragile. Fragmented data limits joined-up services. And if we’re too cautious to use AI responsibly, people will use it anyway — without the guidance, safeguards and common standards that make it safe.

So this is a design challenge.

And design challenges can be solved.

The good news is that we don’t need to overcome every structural challenge before we begin. 

There are practical choices available to us now — choices about where we focus, how we test, what we reuse and how we lead.

Point one: start with a problem worth solving

That brings me to my first point: start with a problem worth solving.

To explain it, I want to take you to back in time to a lunch table at Bell Labs.

Richard Hamming was a mathematician at Bell Labs, an institution famous for discovery. Over the past century, its scientists have received hundreds of prestigious prizes, including 10 Nobel Prizes and 5 Turing Awards. 

Hamming had a habit that made him quietly unpopular. At lunch, he would ask the scientists around him: what are the important problems in your field?

They would answer. Then he would ask a second question: and what are you working on?

And then came the third, the one that really put people off their desserts: if what you are working on is not one of those important problems, why are you working on it?

Hamming said that after that, he was no longer welcome at the table. He had to start eating with the engineers.

I love that story. The people at the table were brilliant, busy people doing highly competent work. They just hadn’t stopped to ask whether it mattered.

Hamming’s conclusion was blunt: if you don’t work on important problems, you can’t expect to do important work. Not because you lack talent, but because talent goes where you point it. His broader point was that exceptional work is not simply a matter of luck or rare genius. It comes from deliberately choosing what deserves your attention, organising your time around it, and having the courage to stay with it when the answer is not obvious.

AI is making Hamming’s question more important. For a long time, much of the heavy lifting was in drafting the brief, building the prototype, analysing the data.

But now with AI – a first draft, a working prototype or a first-pass analysis takes a fraction of the time it once did.

Which means Hamming’s question stops being a lunchtime provocation and becomes the main event. 

As producing becomes easier, choosing well becomes more valuable.

The people at that Bell Labs table carry a warning. They were not choosing their tasks carelessly. They were doing what was familiar and rewarded. 

AI will make familiar and rewarded work faster than ever. If we are careless, we will produce an avalanche of competent work on challenges that weren’t our top priorities.

That’s why the best examples I see start with a real problem. Not a technology looking for a use case. Not a strategy looking for an audience. A problem experienced by a person, a team, or a community.

Take the Department of Veterans’ Affairs. It built a secure AI chatbot in-house to help staff who support veterans find answers specific to the relevant legislation. What I like is the starting point. Not “what can a chatbot do?” but “what do our people struggle to find?”

Here’s a test you can use. Can you describe your problem in one sentence, without mentioning the technology? If you can’t, you may not have a problem yet – you may have a solution looking for one.

And a second question: if we solved this, who would notice?

That doesn’t mean every problem has to be the biggest one in government. Think about where people are getting stuck and what’s at stake if nothing changes. Then ask: can you and your team make a difference?

Point two: make the smallest useful move

Once we’ve chosen a problem worth solving, my second point is make the smallest, useful move.

Don’t wait for perfect conditions. Test something small, with clear boundaries and a clear picture of what success looks like.

But design the poc so that, if it works, it has somewhere to go.

That is where promising work can lose momentum. Across government, there’s no shortage of promising AI experiments. The harder questions come after. Who owns it once the project team moves on? How does it connect to existing systems? What happens when the model changes? And how will we know it’s working: hours saved, better decisions or faster service?

Those aren’t reasons to stop. They’re the work.

The International Labour Organization has examined why strong productivity gains from AI at the level of individual tasks haven’t yet shown up in firm, sector or economy-wide statistics. Part of its answer is that scaling those gains depends on reorganising work and building skills, not just on the tool.

In other words, the pilot isn’t the finish line. It’s the start of the real work.

And stopping is part of the job, too. If a small test tells us an idea won’t work, and tells us early and cheaply, that’s a result. It isn’t an embarrassment.

Hamming described a useful tension here. You need enough belief in an idea to pursue it, and enough doubt to recognise when the evidence is telling you to change direction. That is not hesitation. It is disciplined confidence.

Point three: reuse before you reinvent

My third point is this: reuse before you reinvent. We don’t always need to start from scratch.

We tend to picture innovation as building something new. In government, some of the best innovation is reusing something proven.

Agency-by-agency capability has taken us a long way. 

It can also give us: duplicate platforms, fragmented data and several teams solving variations of the same problem. 

Under fiscal pressure, that matters. Every dollar spent rebuilding something that already exists is a dollar we can’t spend on a problem nobody has solved yet.

Reuse doesn’t mean forcing every agency into the same solution. Context matters. But it does mean asking better questions before we build. Has another agency done this? Could we adapt it? And are we designing so that someone else can use it later?

The National Library of Australia shows what that looks like. Its Trove Metadata Manager, recognised at the APS Data Awards, helps libraries, galleries and museums manage and share their data, and supports sharing across more than 900 organisations. Built once, used by many.

That is not innovation by imitation. It is innovation by accumulation — each team starting from a stronger foundation than the last.

The same goes for knowledge. A lesson shared early can save another team months of work. A failure explained honestly can prevent an expensive repeat. That’s why shared foundations matter. GovAI, the whole-of-government AI service, gives public servants shared training and common tools to learn, trial and apply generative AI. Standards and guidance aren’t glamorous. But they’re how scale becomes possible.

Point four: make it safe to go first

My fourth and final point is about permission: make it safe to go first.

There’s a worry that governance and innovation pull in opposite directions. I don’t think that’s right. The real task is to design governance that helps us move quickly in the right direction — with clear accountability, proportionate assurance, and the ability to correct course.

Good governance should help us move with confidence, not simply tell us when to stop. Clear guardrails give people the confidence to move. 

When the AI Plan was launched, Minister Gallagher described trust, from public servants and the public alike, as a licence for governments to operate. I’d add that trust isn’t only something innovation consumes. Done well, innovation can build it.

And let’s be honest about risk in both directions. There’s risk in trying something new. There’s also risk in keeping an ageing system running, asking people to maintain manual workarounds, or letting a useful pilot die because the funding never arrived.

We’re very good at documenting the first kind of risk. We need to get just as good at seeing the second.

That’s where leaders come in. If you lead a team, you may feel the pull to wait for the right moment, the perfect evidence or the complete plan. But one of your most powerful acts is to make it safe for other people to go first.

Be clear about the outcome and the boundaries. Ask for a small test and real evidence. Recognise a well-run experiment as valuable when it gives us clear evidence — including evidence that we should not proceed. And when something works, don’t leave the team to fight its way into business-as-usual alone.

Closing: one thing

Let me finish by bringing those 4 points together.

Start with a problem worth solving. Make the smallest, useful move. Reuse before you reinvent. And make it safe to go first.

Before I make my ask, I want to come back to Hamming’s lunch table, because his question decides whether everything else I have talked about is worth doing.

What is the most important challenge in your area — and have you organised your time, your people and your technology around it?

Hamming wasn’t just asking people to name the important challenges. He was asking them to choose one and work on it. We’re not short of those in government. Resources are tight, expectations are rising, and many of the challenges people face don’t fit neatly within one agency.

Technology gives us more ways to tackle them. But we still have to decide what matters most, and make room to work on it.

So here’s my ask. It’s small, and it’s three parts.

First, remove one obstacle. Over the next year, identify one thing in your own work that slows a good idea down. Not the biggest thing. Not the thing that needs a decade of reform. One thing within your reach: an approval that adds time but no longer adds assurance, a data set nobody can find, or a pilot that needs an owner. Fix it – or find and support the person who can.

Second, share one lesson. Share the template, the standard, the evidence — and especially what didn’t work. Every lesson we keep to ourselves is one another team may have to learn again.

And the final one, back one person who tries. When a colleague, especially someone early in their career, tries something thoughtful and new, say so. Say it in the meeting. Say it to their manager. It’s much easier to try something new when you know people will back you, even if it doesn’t go to plan. You can be one of those people. That support can change what a colleague is willing to try. 

None of this requires heroics.

It requires ordinary people doing slightly brave things, repeatedly — and noticing each other doing them.

The public sector has always adapted — because public servants have always found ways to meet the moment. Technology gives us new ways to do that — and to do it faster.

Let’s make sure we’re using that capability to solve the problems that matter most.

Remember: Direction matters more than acceleration.

Thank you.