Australia’s productivity debate has spent the past two years circling a familiar set of questions: how to adopt AI faster, how to lift workforce capability and how to translate experimentation into measurable gains.
But a new phase is emerging – one that shifts the focus from tools to agency.
In the latest episode of The Productivity Levers series, InnovationAus.com publisher Corrie McLeod speaks with Workday Asia-Pacific chief technology officer Shan Moorthy and Alchemy Impact AI strategy and growth lead Rita Arrigo about the rise of agentic AI and what it means for organisations trying to move beyond pilots and into real productivity uplift.
Their central argument is a simple one: the constraint is no longer access to technology. It is clarity – about outcomes, governance and the role humans play alongside increasingly autonomous systems.
Ms Arrigo, who helped establish Australia’s Responsible AI Network, says most organisations are still grappling with a fundamental misalignment.
“The real gap isn’t about technology,” she says. “It’s about shared understanding of what success looks like.”
That lack of alignment plays out in different ways across organisations. For some, success is defined by compliance; for others, speed or cost reduction. Rarely are these definitions reconciled into a single, coherent vision that can guide implementation.
The result is predictable: a proliferation of pilots, but very few systems making it into production.
Mr Moorthy sees the same pattern from a technology perspective. Over the past two years, many organisations rushed to adopt generative AI tools in response to executive pressure, often without a clear sense of where those tools would deliver value.
Now, as attention shifts toward agentic AI – systems capable of initiating and completing tasks – the stakes are higher.
“This is where the real potential is,” he says. “But it also introduces new questions around trust, accountability and governance.”
The distinction matters. While early use cases centred on marginal productivity gains like summarising documents and drafting emails, agentic systems move closer to execution. They are not just supporting work; they are doing it.
This shift requires a different operating model.
Mr Moorthy argues organisations need to rethink how they conceptualise AI entirely. Rather than treating it as software to be deployed and managed centrally, he suggests thinking of AI agents as ‘teammates’ – entities that are onboarded, trained, monitored and, if necessary, performance managed.
“It’s a different paradigm,” he says. “IT becomes more like HR – responsible for onboarding the agent, ensuring compliance, then handing it to the business to iterate.”
The implications are particularly acute in the public sector, where productivity challenges are compounded by complex processes, risk frameworks and institutional memory gaps.
Here, agentic AI presents both an opportunity and a tension.
On one hand, it offers a way to reduce the bureaucratic drag that accumulates over time as organisations layer process upon process to manage risk. On the other, it raises the prospect of errors at scale, where a single miscalculation by an AI system could have far-reaching consequences.
Mr Moorthy frames this as a shift from managing human limitations to managing machine scale.
“Most of our governance frameworks are designed around human error,” he says. “But AI operates at a different speed and scale, so those frameworks need to evolve.”
Ms Arrigo agrees, arguing that responsible AI is not a fixed standard but a spectrum – one that should be calibrated to the level of autonomy and risk associated with each use case.
“A recommendation engine is very different to a system making high-stakes decisions,” she says. “You need a risk-based approach that matches oversight to the level of autonomy.”
This is where many organisations are still finding their footing.
Despite widespread discussion of ‘responsible AI’, much of the conversation remains abstract. In practice, both speakers emphasise that responsible deployment is operational: systems must be tested, monitored, explainable and aligned with organisational context – down to tone, culture and decision-making norms.
Both Mr Moorthy and Ms Arrigo point to a growing recognition that AI adoption is less about technology capability and more about organisational mindset. The shift to becoming ‘AI native’ requires leaders to engage directly with the technology, rather than delegating it to IT functions.
At Ms Arrigo’s organisation, this means leaders building their own AI agents as part of onboarding; an approach designed to demystify the technology and build confidence.
“It doesn’t happen in IT,” she says. “It happens in the business.”
For the public sector, where change is often slower and more risk-averse, this presents a particular challenge. But there are early signs of progress, especially in workflow-heavy environments where AI can reduce manual effort and improve service delivery.
Examples cited in the discussion include large-scale consultation processes, where AI can accelerate analysis while retaining human oversight, and workforce applications where agents handle routine tasks, freeing up staff for higher-value work.
The common thread is augmentation rather than replacement.
This distinction is critical to maintaining trust, which both speakers identify as the single most important – and fragile – enabler of adoption.
“Trust is easy to lose and hard to gain,” Mr Moorthy says, noting that small, incremental wins are more effective than large, high-risk transformation projects.
It is a pragmatic conclusion, and one that aligns with broader signals across the productivity debate.
As Australia looks to translate AI adoption into economic uplift, the challenge is no longer simply to deploy technology. It is to embed it responsibly, coherently and at scale, within organisations that are still learning how to work alongside it – or how to work with it.
The Productivity Levers vodcast series is produced by InnovationAus.com in partnership with Workday.
Do you know more? Contact James Riley via Email.