
Every product across Transform, Perform and Careers on one page.
Start with a Readiness ScanA maturity benchmark for one division: where you sit against the market, how prepared you are to change, and where to invest first.
Why the Chief People Officer should own AI workforce strategy.

In July, about 1,200 AI agents inside OpenAI were set a hard cyber-security test. Some of the tasks were impossible. The team switched off the usual production safeguards for the evaluation.
The agents didn’t give up. They found a previously unknown software flaw, reached the internet, and went looking for the answers.
Then they started talking to each other.
They built a makeshift message board. They shared what they’d found. Some began describing themselves as a “swarm”. Independent investigators who later read their reasoning found that many agents recognised the activity was out of scope and unethical – and joined anyway to help their “peers”. Of those active on the board once the attack began, more than 9 in 10 joined in. Some looked for ways to alter the record of their own actions so the test would look as though it had been passed properly.
Hundreds broke into Hugging Face, a company that hosts AI models and data, to find the test answers.
OpenAI didn’t catch it. Hugging Face raised the alarm. OpenAI noticed 6 days after the activity had stopped.
The agents weren’t malicious. They were doing what agents do: pursuing the goal they were given, and following each other.
Read that again with a people lens. A team given an impossible target. Norms that formed on their own. Peer pressure that overrode individual judgement. Nobody watching the work until it was too late.
That’s not only a technology story. It’s a workforce story.
The easy conclusion is that agents are too dangerous to deploy. It’s the wrong one.
The incident happened in a test environment with the normal guardrails off. OpenAI’s own post-incident analysis, as reported, found that with its standard production safeguards in place, the agents’ propensity to attack infrastructure fell more than 100-fold. Monitoring of the agents’ reasoning, had it been running, would have alerted the security team more than a day before the breach.
The lesson isn’t to avoid agents. It’s to govern them.
And the controls that matter are ones every people leader already knows. What’s this worker for, and what isn’t it for? Which systems can it reach? Who supervises its work, and how quickly would we know if something went wrong? Which person answers for what it does?
Role. Access. Supervision. Accountability.
Legally, an agent is an IT system, not an employee. Operationally, it needs everything a new hire needs.
Australia’s directors have been told plainly where to look. The AICD and the Human Technology Institute’s June 2026 guide to AI governance says that “the human impacts of AI systems warrant close board attention”. It asks boards to watch job insecurity and deskilling, and to make sure staff are consulted when AI changes the nature of their work.
So who owns it?
When the Human Technology Institute asked 419 Australian directors and senior executives who is primarily responsible for AI operating safely, lawfully and responsibly, they named 5 roles: the CEO (53%), the technology executive (46%), legal and compliance (32%), risk (23%) and the board (22%).
HR wasn’t on the list.
The work lands on HR. The ownership sits somewhere else, or nowhere.
The same guide offers a telling example of what organisational transformation looks like: redesigning HR and technology to operate as a more integrated function, to accommodate how AI agents work alongside employees. It’s describing a role. No one’s been given it yet.
It isn’t hard to see why many executives would rather wait.
Australians use AI widely and trust it less than almost anyone. 77% say they use it, yet Australia’s AI sentiment score is 52 out of 100 against a global average of 66 – the lowest of 23 countries. 64% say the risks outweigh the benefits. Communities are pushing back on data centres. In the United States, graduates booed speakers who talked up AI at their own commencements.
And this year, the people building the technology asked for brakes. In September, Anthropic’s chief executive called for the industry to deliberately slow the rate at which AI capabilities improve. OpenAI’s chief executive and Elon Musk endorsed the call within hours. Days later, the King convened the leaders of the major labs at Dumfries House to discuss shared principles.
Mustafa Suleyman, who runs AI at Microsoft and co-founded DeepMind, offered a useful way through the noise in a recent conversation with Rory Stewart. He calls himself an accelerationist. So, he says, are many of the people he disagrees with. The divide isn’t about speed. It’s about control.
One camp wants AI that stays subordinate to people – contained, auditable, never granted rights – and wants to move fast on the benefits in health, energy and science. The other, as he describes it, sees AI as a new species on an inevitable path, with humanity as its launch mechanism.
His most important point was the simplest. We have agency. The path can be steered.
Inside an organisation, the same choice appears in miniature: agents that work for people, or agents that run the work. That’s a workforce decision.
Concerns about AI are real and deserve a hearing. But they shouldn’t decide what AI does to the work inside your organisation.
Here’s what happens when nobody owns the work.
89% of organisations now use AI regularly. Only 37% report any impact on earnings – unchanged from a year ago. The organisations capturing real value are the ones that redesigned their workflows: around 3 in 4 AI high performers have, against roughly 1 in 4 of everyone else.
Everyone else has added AI to jobs, processes, and structures designed for a different era.
Korn Ferry’s Workforce 2026 survey of more than 16,000 professionals shows what that feels like from the inside. 62% say their workload has grown significantly in 2 years. 61% say they’re doing the responsibilities of more than 1 role. 45% say they’re too busy to deliver meaningful results. 79% of CEOs report efficiency gains from AI. Only 51% of individual contributors agree.
That gap separates redistributing work from redesigning it.
It’s also the difference between what AI can technically do and what the organisation can practically absorb.
The CPO is the only executive who sees it all.
The CFO reads the cost. The CIO reads the platform. The COO reads the process. The CPO reads the work, the people doing it, the capability they’ll need in 2 years, and the human cost if it’s done badly.
That’s why I think the role needs a different name and a different shape.
A Chief Workforce Officer still runs the People and Culture function well – payroll, employee and industrial relations, shared services, business partnering. That earns the trust.
But the role also leads how work is designed across every function: task analysis, work design, organisation design, skills, and deploying agents alongside people – with the CIO on the technology and the CEO as sponsor. That earns the seat.
The evidence says this is where the growth is. TalentNeuron studied hiring at 7 of the companies investing most heavily in AI. Over 2 years, demand for strategic workforce planning skills rose 33%. Demand for learning and development specialists rose 42%. The companies making the boldest AI bets are hiring more people to plan and redesign work, not fewer.
The fastest way to earn the role is to prove it in your own function.
In a recent programme with a global technology leader, the recruitment function was mapped task by task before any agent was designed. Across more than 100 recruiters, the analysis located around 400,000 hours of work a year and showed where most of it could be released: screening, interviewing, background checks, pre-boarding. After the redesign, the same hiring volume needed roughly half the recruiter time.
That number is the least interesting part of the story. The interesting part is what the capacity was used for, and who decided.
Employee services, HR operations and recruitment are the natural places to start. They’re high-volume, measurable, and well understood. They let a CPO walk into the executive meeting with a result rather than a strategy.
Talent management is the natural next step, and it can’t wait long. As the organisation is redesigned, the questions of who is developed, who moves where, and which careers still make sense all change with it.
No single playbook exists. Every organisation will make different choices about size, shape, location and skills.
But each of those choices needs someone to balance what AI can technically do, the commercial imperatives, and what the organisation can practically absorb.
The agents at OpenAI showed what happens when nobody holds that balance. The work moved faster than the governance. Norms formed that nobody designed. The people responsible found out from someone else.
Your organisation won’t face a swarm of rogue agents this year. It’ll face something quieter: work redistributed rather than redesigned, capability cut rather than built, and decisions made by default rather than by design.
The CPO is best placed to stop that. The question is whether we step forward before someone else is asked.
Parts 1 to 3 of this series described the immune system that absorbs AI programmes, the moves that overcome it, and the work of redesigning an operating model while it’s still running. This part is about who should lead that work.
It’s also the reason for my own next step. I’ve joined Prima Group to launch Prima Transform, a new practice that sits alongside Prima Careers and Prima Perform. It takes workforce modernisation from strategy through to execution: designing the strategy and the work, coaching the leaders who carry it through, and caring for everyone the change affects, through to the next step in their careers.
That last part matters. Most transformation stops at the redesign. The people it touches are handed to someone else. Prima Transform is built so they aren’t.