Why scaling AI is a human problem long before it is a technical one.
Most organisations now have AI tools. Far fewer are getting real value from them, and the gap is rarely about the technology. It is about the human side: leadership, culture, change and clear decisions from the top.
That was the heart of Jon Whittle's opening keynote at our AI Leadership Summit. Jon has spent three decades in AI, from NASA to leading CSIRO's Data61 and founding Australia's National AI Centre, and much of this draws on his new book, AI for Business: A Guide to AI Adoption. Better still, he backed it with evidence, including a Stanford study of 51 successful at-scale AI deployments.
2025 was the year of pilots. Companies rolled out licenses and ran a few experiments, assuming that if you gave people the tools, good things would follow. This year the question has changed: how do we go from a few pilots to real, enterprise-wide return?
To work out where an organisation sits, Jon uses a simple test. Every business is an AI watcher, an adapter or a disruptor. Watchers make small investments and keep an eye on competitors. Adapters invest more, but mostly to speed up what they already do. Disruptors use AI to do genuinely new things. On his own count, around 70 to 75% of Australian organisations are still watchers. None of the three is wrong, but you have to choose deliberately, because if you do not, different people assume different answers and you never get alignment.
Jon's sharpest material was his five "anti-patterns", the mistakes he sees again and again.
The efficiency trap is treating AI purely as a way to go faster. You miss the bigger opportunity to rethink how the business works, and as he puts it, if you apply AI to a broken process, all you get is a faster broken process.
Use case chaos is when leadership says "we must do AI" and the organisation collects every possible use case. He has seen companies gather 600, hand the pile to IT to prioritise, and by the time it is filtered to a workable few, the shortlist is already out of date.
The ambition gap is the failure underneath both: leadership never sat down to decide what they are actually trying to achieve.
Governance paralysis is the opposite problem, moving too slowly. Teams decide they must reinvent their entire governance model before starting, so they delay. Usually the structures you have can be adapted rather than rebuilt.
And who's in charge is the quiet killer. Most organisations default AI to the CIO or CTO, which is fine for a watcher but wrong for a disruptor, who needs someone with change and people skills reporting to the CEO. Worse is when no one has decided. Jon asked one leadership team who was accountable, and after an awkward pause someone offered, "I guess we all are." Which means, when something breaks, no one is.
Jon's constructive answer is a set of 14 dimensions across leadership, people and governance. The unifying idea is that the barriers are human, not technical, which is good news, because these are skills most leaders already have.
A few points landed hardest. On cost, most business cases only count the technology, but his favourite stat, from Stanford, is that for every dollar you spend on the tech, you should budget around ten for the change management, training and people around it. On method, every successful deployment Stanford studied was iterative, and 60% failed at least once before they worked, so treat early failure as information, not a reason to pull the plug. On oversight, checking every AI output just moves the work and bores the checker, so take a risk-based approach: successful companies let AI run about 80% of the time and reserve human eyes for the 20% that matters.
And on productivity, personal gains are real, 5 to 75% depending on the task, but they do not automatically become organisation-wide gains. Saved time gets swallowed by more meetings and noise unless you plan for it. Running through all of it is a shift in who owns this: every leader is now an AI leader, not just the technical ones, which means leading with real care, because the fear and overwhelm your people feel are real too.
Jon's throughline is simple: the technology is the easy part. The organisations pulling ahead are not the ones with the most pilots or the flashiest tools. They are the ones being deliberate about their ambition, honest about the real cost of change, and willing to iterate their way to value. For leaders staring down the hype, that is oddly reassuring. Most of the skills you need, you already have.
With thanks to Jon Whittle, founder of Goldilocks AI and professor at Melbourne Business School, for opening The Leadership Institute's Leading AI Innovation Summit. His book, AI for Business: A Guide to AI Adoption, is out now through CSIRO Publishing.
More information about the summit