An agent can process a thousand routine payroll cases without trouble, but the real test comes with the unusual one: conflicting records, a rule applied wrongly, or an answer that looks right and isn’t. At that point, three questions matter: who notices, who steps in and who has the authority to fix it?Most companies decide what AI agents will do long before anyone asks what that means for the people doing the work. A typical example: the CIO presents a roadmap for AI agents in payroll, recruiting and scheduling, measured in hours saved. The CHRO asks which roles will change, which skills will matter and who handles the cases the agent can’t resolve. Both questions are fair but by the time HR asks its question, the main design decisions have usually already been made.
That timing has a cost. According to Gartner’s HR Symposium/Xpo London highlights, only 38% of CHROs and CIOs share an understanding of how AI will affect the future of work, and only 14% of HR agentic AI initiatives are rated fully or highly successful. When the two key leaders responsible for people and technology picture different outcomes, the organisation ends up running two projects under one name.
The answer isn’t for HR to become a technology function. It’s for HR to come to the first planning meeting with what IT can’t supply on its own: a clear picture of how jobs, decisions and accountability will work once an agent takes on part of the load. Deloitte Deutschland makes the same case, arguing that CIOs, CHROs, risk officers and business leaders have to act as one connected leadership system. HR leaders who wait to be invited in end up managing consequences, whereas those who join early help decide what the consequences are.
What the Gartner Numbers Really Say
Gartner’s London findings include three figures worth reading carefully:
- 38% of CHROs and CIOs share an understanding of how AI affects work.
- 14% of HR agentic AI initiatives are fully or highly successful.
- 56% of HR leaders name poor data quality as the biggest barrier to AI agent success.
The 14% is easy to misread. It doesn’t mean 86% of projects failed. It counts only the projects in the top success categories, and many of the others are simply at an early stage. The research also doesn’t prove that misalignment between leaders causes weak results.
Over time, though, the 38% is the more telling number, because it won’t improve on its own. Early projects mature. A disagreement about what AI is for tends to stay until someone deals with it. Germany shows a similar gap between adoption and control: in Deloitte’s AI study, 27% of executives said they already use agentic AI at least moderately, while only 19% have mature governance in place.
That raises an awkward question. If leaders haven’t agreed on what an AI initiative should change, how will anyone know whether it worked?
Alignment Starts With the Work
Sitting in the same steering meetings isn’t the same as being aligned. The real test is whether HR and IT can describe, in plain terms, how a job works once an agent has taken over part of it.
A technology roadmap says which tasks the agent will handle. A workforce roadmap has to answer the harder questions:
- Which decisions stay with people?
- What’s left of a role once the routine work goes?
- Which skills become more valuable, and which matter less?
- Where does a person need to check the output?
- Who is responsible when the agent gets it wrong?
HR Performance puts it in practical terms: HR and IT need to settle who owns the business side, who owns the technical side, who decides when things are unclear and who has the right to stop an agentic process.
Gartner’s CHRO priorities for 2027 show why this can’t wait. AI agents systematically hand ambiguity, judgment calls and recovery work back to employees. Yet, as it-daily.net reports, only 24% of CHROs say their leadership team agrees on who is accountable for AI agents.
This is why WiseForce Advisors describes today’s CHRO as the chief translator between the AI agenda and the workforce, a role that now looks more like systems design than traditional HR.
What HR Brings to the AI Conversation
HR doesn’t need technical expertise to shape the AI agenda. What it brings is the ability to say what changes for people, and to say it before the build starts rather than after.
If an agent takes over routine admin, what’s left of the role? Does the job lean more on judgment? Will employees need stronger diagnostic skills? Will managers spend more of their week on escalations? Haufe calls this finding the right human-agent ratio: deciding which tasks go to agents, which stay with people and how resources are planned around that split.
These answers don’t need to be polished. A CHRO who arrives at the first planning session with a rough outline of how three or four key roles will change sets the terms of the discussion. Without that outline, the CIO’s task list becomes the default plan, and HR inherits outcomes it didn’t design.
Data quality belongs in the same conversation. A contributor to WiseForce Advisors’ Experience Edge series notes that processes as written down are often not the processes people actually follow, and AI makes that gap bigger. HR often knows best where the real process differs from the official one. Bringing that knowledge in early won’t fix bad data, but it does show IT where to look.
Settle Accountability Before Launch
An agent can process a thousand routine payroll cases without trouble, but the real test comes with the unusual one: conflicting records, a rule applied wrongly, or an answer that looks right and isn’t. At that point, three questions matter: who notices, who steps in and who has the authority to fix it?
A useful CHRO–CIO session starts with one question: what happens on the day the agent is confidently wrong?
Working through a few real scenarios usually brings the gaps to the surface quickly. From there, leaders can agree on five things:
- Decision rights: which decisions the agent can carry out alone and which need human sign-off.
- Exception handling: where unresolved, disputed or high-risk cases go.
- Intervention authority: who can pause the process, override an output or ask for a review.
- Capacity: how much time the people handling exceptions actually have for it.
- Success measures: how quality, correction effort and service are tracked alongside time saved.
Haufe Akademie describes this as the human-in-the-loop principle: final responsibility for results stays with people, backed by clear rules on which decisions agents may make alone. Naming an owner is the first step. That person also needs the information, the authority and the hours to do the job properly.
A Career Opportunity for CHROs in Transition
For CHROs and senior HR leaders weighing their next executive or advisory role, this gap gives them something concrete to talk about.
“Led AI transformation” says very little. “Partnered with the CIO to define human decision rights and exception ownership for an AI-enabled payroll process” says a great deal, as long as it’s true. Boards and CEOs hear plenty about AI from vendors and consultants. They hear far less about who owns the decision when the system fails.
The strongest examples are short: the problem, what the leader did and what changed as a result. That kind of judgment is also what boards look for in non-executive and advisory roles, as WiseForce Advisors explores in its guide to building a portfolio executive career. It matters even more now that, as WiseForce Advisors argues, restructuring has become the operating model rather than a phase companies pass through.
Reinvention, Not Retreat
The 14% success rate will rise as projects mature. The 38% will only rise if someone decides to close the gap, and HR is well placed to do it. The CHRO doesn’t have to own the technology, and the CIO can’t be expected to resolve every workforce question. Between them, they need to connect what the technology can do with clear decision rights, workable roles and human judgment. HR leaders who take that on from the first meeting will find their advice sought well beyond HR.
