AI Is Projected to Cut Core HR Headcount by 30% and Turn the Rest Into Agent Supervisors

Every CHRO planning a leaner function for next year is being handed two goals that work against each other. AI agents are expected to let companies run core HR with roughly 30% fewer people. The team that survives the cut is expected to onboard those agents, train them, and answer for what they do.

Cutting a third of the function assumes that whoever remains can supervise a workforce of agents. Almost no HR team is staffed to do that today, and the skills it takes are not the skills being automated away. Companies that cut first and build the capability later will hit their headcount number and lose control of the work.

Coordinators become supervisors

The 30% figure comes from The Josh Bersin Company, an HR advisory firm whose 2026 research maps more than 100 agent applications across the function. The firm is clear that this is not a straight cut. Roles disappear, others take their place, and more than 30 new job titles have already appeared across HR and IT.

What agents take over is the transactional work, which is where most HR headcount currently sits. What is left behind is oversight. That is a different job than processing benefits enrollments, and the person doing it needs different training and different instincts.

The budget is where this gets decided. A 30% cut only saves money if the roles replacing the old ones cost less, and no one has shown that they will. Bersin's own framing is that the work changes rather than disappears, which leaves CHROs with a question they can answer internally before they answer it externally. Who on the current team can move into the new job?

Competing with engineers

Hiring for this is harder than it looks, because the job HR needs is not one the market currently sells. There is no established pipeline of agent operations leads, no credential that certifies one, and no salary band a compensation team can benchmark against. A CHRO who opens a requisition is guessing at the title, the level, and the pay.

The adjacent market shows what that guess is up against. A forecast from the recruitment firm Robert Walters and the global employment platform Native Teams puts US demand for AI professionals more than 800,000 above what the domestic market will produce by 2028, which would leave roughly one in two of those roles unfilled. Engineers and data specialists are not HR people, but they are bidding for the same scarce understanding of how these systems behave, and they are doing it with far larger budgets. Anyone HR brings into this work gets priced against that market whether or not the job title says so.

Meanwhile, 13% of organizations are already planning to reduce HR headcount by an average of 22%. They are trimming a function to fund a role that has no definition and a price set by a market they do not compete in. Both firms conclude that the organizations moving fastest on AI are the ones that can reach specialist talent regardless of where it sits, which is a self-interested read from two businesses that sell exactly that, and also the most plausible answer on the table.

Still mostly pilots

If the capability were easy to buy, the deployments would already be visible. They are not. Over the past year, use of AI across HR rose by somewhere between zero and six percentage points depending on the area, and most companies are still running pilots. McKinsey's HR Monitor 2026 finds the live deployments concentrated in routine administrative work, which is the cheapest thing to automate and well short of what the 30% forecast assumes.

McKinsey puts the slow progress down to messy technology and a shortage of internal skills, and prescribes a new operating model with cleaner data underneath it. The technology is available now, and what most organizations lack is the structure to run it at scale.

Liability needs a name

Accountability has no obvious owner in any of this. An agent that misapplies a leave policy or opens employee records that it should not touch creates an exposure. Matt Prebble, chief executive of Accenture UK and Ireland, told the Financial Times that HR directors may end up onboarding and training these agents, and that companies may need new executive roles built around trust and governance. Somebody has to answer for a worker who cannot be fired or coached.

The sequence matters more than the timeline. Decide who owns an agent before it goes live, settle who is qualified to supervise it, and then talk about what the function costs. Companies that run that order backward will hit the 30% on schedule and find they have cut the people who were supposed to be watching.