One AI-savvy architect can cut costs and free staff for human work – without a top-down mandate
Human resources (HR) leaders are increasingly turning away from top-down artificial intelligence (AI) mandates in favour of appointing a single internal "AI architect" – a specialist who absorbs the risk and complexity of deployment so the rest of the workforce can focus on people-facing work.
That is the view of two industry figures. Josh Campbell, head of people and capability at Tritium, and Pete Bradd, chief evangelist JAPAC at Miro, both argue that forcing staff to adopt unfamiliar tools produces poor outcomes and reputational risk
A single trusted expert can quietly automate routine HR work without disrupting how the rest of the business already operates.
Why AI mandates fall flat with HR practitioners
Campbell, who has spent more than four years working at the intersection of HR and AI – including a stint on Kmart Australia's chatbot project with recruitment technology company Sapia – said mandating tools for staff who don't understand them creates more risk than it removes.
"I don't want to force them to use technology that they don't understand ... or enjoy using," he said, adding that his approach is instead "to solve problems for them so that it's transparent in their mind and they don't have to learn how to use anything."
Bradd made a similar point from the vendor side, warning that blanket policies misjudge how people actually behave. Citing recent workforce research from AI proficiency analytics firm Section, Bradd said employee proficiency has barely moved even as usage mandates spread, estimating it sat at roughly 8 per cent in July 2026, up only slightly from 3 per cent in January.
He added that close to three in 10 workers remain "AI novices" who are more likely to default to free, unsanctioned consumer tools that carry none of an employer's enterprise guardrails.
The case for a single AI architect
Rather than train every HR business partner to use generative tools, Campbell has built an internal, agentic AI system – software made up of multiple AI agents that complete multi-step tasks with limited human input – covering a learning platform, an intranet and a suite of workplace tools.
It includes a learning management system that self-generated more than 100 courses customised to Tritium's own policy documents and to local legislation where his teams operate throughout the world.
He estimated the total build cost at roughly $300 for the AI tools, and a month of work, replacing licensed learning management and parts of the human resources management software the business previously paid multiple tens of thousands of dollars a year to use.
“I don’t want to force people into using technology they may not feel comfortable with. People work differently and have different abilities, so the focus should be on providing useful options and support rather than mandating one approach. Campbell said of his reluctance to mandate use, arguing instead for "freedom within a framework."
Bradd agreed that governance, not mandates, should carry the weight of managing risk. "How do you stop the risk from happening?" he said. "That, to me, should be the job of HR leaders and technologists," pointing to enterprise features such as prompt redaction, which strip sensitive company or customer data before it ever reaches an external large language model.
Weighing the risk against the reward
Neither expert dismissed the risk of concentrating AI knowledge in one person. Campbell said he commissioned an AI-driven security audit of his internally built systems – a process that reportedly ran more than 400 agents over eight hours – before handing the platforms to Tritium's internal IT and security teams for further testing.
"I absolutely recognise there's going to be that risk," he said, noting he holds a psychology degree rather than a formal software background, even though he is using "exactly the same tools" as professional developers.
For Bradd, the upside is what happens once routine load is lifted from HR teams. Individuals may already be far more productive with AI, he said, but that has not yet translated into company-wide productivity gains, because most modern work depends on collaboration rather than solo output.
Freeing HR business partners from administrative load, both argued, ultimately buys back time for the part of the job AI cannot replace. HR must lead AI implementation rather than support it, that human element – coaching, relationship-building and judgement calls – is exactly where practitioners say their time is best spent.
It comes as separate research shows AI adoption is increasingly employee-driven rather than led by leadership, and as other organisations find that a culture-first approach succeeds where blanket AI mandates tend to stall.
The Tritium and Miro experience suggests the more effective route for some organisations may not be getting everyone to use AI – but trusting one architect to build it responsibly on everyone else's behalf.
