Accenture’s UK CEO told the FT last week that HR directors may one day be responsible for onboarding and training AI agents. It’s a neat idea, and I can understand why he said it – to gain notice – but the analogy is wrong.
The instinct is understandable; HR already owns onboarding, capability, and workforce governance. So when “digital workers” show up, we reach for the closest analogy and assign it to the same function. But AI agents aren’t employees, and treating them like they are is already part of the problem.
Managing a person is about expectations, development, and accountability. Managing AI in a workflow is about something else entirely: breaking work into tasks, defining decision points, setting thresholds for intervention, and being explicit about who is accountable when outputs are co-produced. For example, if an AI system is triaging cases or drafting responses, the real design work sits in escalation rules, approval boundaries, and failure handling, not in “training the agent”. That is not an HR process but an operating model decision.
We are seeing the same misstep with AI literacy. The default response – “we’ll train everyone” – is comforting and ineffective. The direction of travel in regulation, including the EU AI Act, is toward role-specific understanding grounded in real use and risk. What someone needs to know about AI depends on what they are asking it to do, what decisions they influence, and what happens when it goes wrong. Generic courses will not get you there.
This is where L&D needs to be honest about the change in the way we work.
This is not a demand signal for more AI training but a demand signal to get involved in how work is redesigned. If tasks, workflows, and decisions are being redefined, then capability is being set at the point of design. Arriving afterwards clutching a training programme, e-learning module and support guide is too late.
That changes the L&D role from delivering ‘learning’ to shaping roles and from building content to defining what competent performance looks like when humans and AI share the work.
It’s not about “who manages AI?”
It is “who is accountable for performance when humans no longer work alone?”
If the answer is “HR,” the organisation has misunderstood the problem and if the answer is unclear, the organisation is not ready.