The usual claim is that we are falling behind countries such as the USA and need to move faster. I think that’s only partly true.
Both Australia and New Zealand have a pretty good history of adopting technology quickly when the value is obvious. In the 90s, New Zealand helped make EFTPOS normal at a national scale very early, while more recently, Australia has embraced tap-and-go payments well ahead of many similiar countries.
That is worth remembering because it tells us something important. Australians and New Zealanders are not naturally resistant to technology. We tend to adopt things when they’re useful, trusted and easy to fit into everyday life.
Payments are a good example. The benefit was obvious and the experience was simple. There were risks, of course, but the ecosystem around the technology built trust. Banks, retailers and payment providers made it safe enough for people to trust and the results are now widespread adoption.
I think something similar is happening with AI.
As individuals, we’re not reluctant users of AI at all. Plenty of people I know who have no background in technology are now regular users of ChatGPT. Often it starts as a bit of fun, but pretty quickly they’re using it to answer questions, prepare letters, plan things like holidays and gym workouts or get advice on parenting and dieting
That comfort with AI is naturally making its way into the workplace as well. Employees are experimenting, learning and finding useful ways to save time.
And then they run into the organisation.
Employee: “I built an agent that saves me three hours a week!”
Organisation: “Excellent. We have established an AI Steering Committee to decide whether you’re allowed to use it.”
Sound familiar?
The real gap, in my view, is not individual AI adoption. It’s whether organisations have the confidence, governance and operating model to turn all of that experimentation into something that can operate safely at scale.
From a ServiceNow perspective, that distinction is becoming increasingly important.
The value of AI is already moving beyond drafting an email, summarising a meeting or searching through documents. The more interesting opportunity is AI acting across real enterprise workflows using the right data, permissions, approvals and controls.
Once that happens, the conversation becomes much less about the model and much more about the system of work around it.
To build trust we must focus the layer of software that enables a large language model to interact with data and tools to perform tasks. This layer of software is called the Agentic AI Harness. Frontier models regularly change, and they can often be swapped out for other models. Models are probabilistic, the same input will create different outputs and they have an inherent risk of unintended behaviour whether that be from prompt injection or simply bad data fed to the model. The AI harness on the other hand is the more enduring and deterministic component of an agentic system. It is where we can apply proven security practices of least privileges, identity and access management and monitoring.