An AI tutor is easy to demo and hard to put in front of students. At the start of the year, when we set our principles for working with AI, we knew one of the biggest changes would be how quickly we could turn an idea into something real. That has certainly proved true. This week I have been writing those principles down in public, because they are now being tested in the work rather than just agreed in a room. Alongside that, AI has been framed in Australia as a sovereignty question, which is a striking shift in how the country talks about a technology that, until recently, sat in the IT budget.
What we mean by an AI tutor is set out on the AI tutor for schools page.
AI is in our work, and relationships are the foundation
The first principle we defined was: AI is in our work, and relationships are the foundation.
For us this means using AI as part of the work without handing over responsibility for it. It can help us explore ideas, reduce friction and take away some of the drudgery, but people still frame the problem, apply judgement and own the outcome.
The relationships part matters just as much. For the last two weeks we have chosen to work together in the office rather than remotely. Solving these problems side by side has been a good reminder that even as AI becomes more embedded in our work, the human connections around the work still matter enormously.
We spend most of our time helping other teams remove drudgery and make better use of AI. This time we have been getting our own house in order, with the aim of improving the quality and consistency of our work, and giving people more space to focus on the things that need human judgement. The idea of an AI-enabled "factory" is a useful way of thinking about repeatability and scale. What we are actually building feels closer to an operating system for how the team works: something that supports people, creates consistency and removes friction, while keeping them firmly in control.
Prototype an AI tutor fast, and commission it with care
AI now lets us move from an idea to a working prototype incredibly quickly. That is powerful, because it means we can learn sooner, get something in front of people earlier, and improve it based on what actually happens rather than what we think might happen.
Speed creates its own risk. Something that works in a demo is not necessarily something that is ready to rely on. Our second principle was: prototype fast and commission with care.
We want to experiment freely in prototype mode. Try things, learn quickly, and be willing to throw work away. But when something moves into production, the bar changes. Security, reliability, testing, support, governance and clear ownership still matter. AI should make it easier to explore and learn. It should not lower the standards we apply when something becomes part of how people actually work.
The sovereignty inflection
Australia's Defence Minister has called AI "the biggest sovereignty inflection point since the Second World War". It is a striking statement, but it reflects how quickly AI has moved from a technology issue to a question of national capability. The reporting is worth reading for the framing alone.
Australia cannot be satisfied with hosting data centres and buying models built elsewhere. Access to leading systems matters, as do local talent, research, products, and the ability to shape how AI is used. If the window is measured in days, weeks and months rather than years and decades, we need to move with real urgency, and we are late getting moving.
That still leaves the copyright question sitting underneath the sovereignty one. You cannot talk seriously about national capability in AI without talking about the data the models learn from, and who gets to decide.
We were already clicking less
A tongue-in-cheek note to finish, for those worried about their skills atrophying because we just ask AI for answers. We were already at it in 2020, with 65 per cent of Google searches resulting in zero clicks. We have just changed the text entry box we are using.