The future of HR
AI won't fix an HR process no one trusts
Every HCM platform shipped HR agents this year, and they're capable. The CHRO's job in 2026 is deciding which people decisions a machine is allowed to make, and owning that answer.
I've spent a lot of my career on the HR side of large transformations, and there's a moment I've watched play out a dozen times in the last year. A CHRO sits through a demo of the new AI agents in their HCM platform. The demo is good. The agent drafts the performance summary, surfaces the internal candidate and answers the benefits question without a ticket. Heads nod.
Then the meeting ends and everyone goes back to what the demo never touched: a job architecture nobody has cleaned up since the last reorg, a skills "taxonomy" that's three spreadsheets and a shared drive, and a performance history carrying every bias the organization has ever had.
That gap, between what the agent does in the demo and what your data and processes can support, is the whole CHRO agenda for the next 2 years. The software was always going to be the easy part.
Who answers for AI in HR
- Chief HR OfficerDecides which people decisions a machine can make
- HR agentsNew this year on every major platform
- The 2%The cases that needed a human
- Operating modelWhere nearly 30% of the gains come from
- People dataEverything AI learns from your history
- Skills architectureWhat the agents assume is underneath
- Workforce planBigger than this year's headcount
- Entry-level rolesWhere senior talent used to start
- HR's own team30% or more of headcount could shift
Every major platform shipped HR agents that do real work
None of this is speculative. The agents shipped across 2025 and into 2026.
| Platform | Agents | Work they took on |
|---|---|---|
| Workday | Illuminate agents | Recruiting, talent mobility and HR cases |
| SAP SuccessFactors | Joule agents | Performance, career development, payroll questions and HR service |
| Oracle Fusion HCM | More than a dozen agents | Everything from interview scheduling to internal job matching |
| Microsoft | Employee Self-Service agent | Sits in the flow of work, with skills intelligence sold straight through Copilot |
The operational results hold up too. Robinhood reported deflecting 70% of employee HR requests before they ever reached a human. That's roughly 2,200 hours a month of work that simply went away. The City of Raleigh reported a 98% deflection rate. When the work is high volume and rule bound, these agents earn their keep fast.
70%
of employee HR requests deflected before they reached a human at Robinhood
2,200
hours a month of work that simply went away
98%
deflection rate reported by the City of Raleigh
Who handles the 2% that needed a human?
A 98% deflection rate leaves a question nobody asks. Somebody still takes the cases the agent couldn't close, and by definition those are likely to be the hardest ones.
Whether to use AI in HR is settled. What's open is which decisions you hand to a machine, where the person stays in the loop, and who answers for it when it gets someone's livelihood wrong.
Adoption is running well ahead of value
Every CHRO is living inside this tension, even if nobody has named it in your leadership meeting.
About 39% of HR functions have adopted AI somewhere, by SHRM's 2026 count. Gartner surveyed HR leaders and found 88% say their teams haven't yet seen significant business value from the tools they deployed. Even so, more than 80% plan to add agentic AI to their function inside the year.
Most teams haven't gotten value from the AI they already have, and they're speeding up into more of it. What drives that curve is fear of being left behind. The CHROs who slow down just enough to build the foundation first are the ones who end up in the 12% that got value, and what got them there was the dull work underneath the agents.
Gartner also put a number on where the value comes from. Nearly 30% of AI's productivity gains trace to changing the HR operating model itself, apart from the tool and apart from training people on the tool. The license is just the permission slip. The operating model is the work.
Put AI on top of an HR process your people already distrust and it automates the distrust, scales it, and removes the human who used to soften it.
What AI inherits from you
"Garbage in, garbage out" became a cliché because it's true, and in HR the stakes run higher than in finance. When AI learns from your historical people data, it learns all of your history.
The cautionary tale everyone in HR now knows is Amazon's early recruiting model, which taught itself to prefer male candidates because it trained on a decade of resumes from a male-dominated field. That's the default outcome of pointing a learning system at biased history. If your promotion records, pay decisions and performance ratings carry the patterns of the past (and almost every organization's do), an AI trained on them will reproduce those patterns at scale and hand them back as objective. Bias laundered through an algorithm looks like math, which makes it harder to challenge.
That's why an AI initiative in HR stops working the moment you run it as an IT project that HR sponsors. It's an HR accountability problem that IT enables. The difference shows up the first time a rejected candidate, or a regulator, asks you to explain a decision your agent made.
The skills-based org is the architecture AI is waiting for
What HR leaders most want from AI is matching people to open roles, surfacing who's ready for what and targeting reskilling. All of it depends on something most organizations haven't built: a real skills foundation.
Mercer's latest research found fewer than one in five organizations doing skills-based talent management at any real scale, even though the same research keeps showing those organizations are far more likely to hit their workforce outcomes.
The agents your vendor is selling assume a clean skills model underneath. Roughly half of organizations still can't reliably map skills to jobs, and those organizations can't get value from the matching, mobility and learning AI they're being sold, however good the agent is.
Skills architecture used to be a someday project. Now it's the near-term precondition for everything else on the AI roadmap.
Plan past this year's org chart
If you're planning your workforce strategy around current headcount, you may be solving the wrong problem.
AI is killing the path to the first job
The clearest labor-market signal so far is the quiet collapse of the entry-level pipeline. Researchers at Yale found employment for the youngest software developers down about 20% from its 2022 peak, with entry-level postings down by half. Companies are mostly keeping the people they have and closing the door to new ones.
Still thereGone
The World Economic Forum projects heavy churn this decade, with tens of millions of roles created and displaced and a large share of workers needing to reskill.
For a CHRO the implication is sharp. If AI absorbs the routine work junior people used to learn on, where does your next generation of senior talent come from? That's a board-level workforce risk, and you won't see it if you only look at this year's org chart.
Your own house is part of the redesign
HR gets no exemption from what it's deploying. Josh Bersin's research argues that AI agents could automate a large share of core HR processes and shift 30% or more of traditional HR headcount toward higher-value work.
Deloitte found two-thirds of executives believe functions like HR must fundamentally change, and only 7% think they're making progress.
That gap is the CHRO's risk and opportunity at once. Treat it as a cost-cutting story and you lose the narrative, and probably the function's seat at the table. Treat it as a capability shift, with fewer hands on transactions and more on judgment, governance and the human moments a machine can't hold, and you redefine what HR is for.
Somebody is going to decide what HR looks like in 2 years. Better the CHRO than a finance-led automation program.
What the CHRO now owns, and can't delegate
Some of this you assign. The technical build goes to IT, regulatory tracking goes to legal, and change management goes to your business partners. The core can't be handed off.
- The operating-model call. What AI does, what people must do, and exactly where the handoff sits.
- A standing bias audit for every system that touches hiring, promotion, pay or performance.
- The human-in-the-loop policy for consequential decisions, written down and enforced. An agent can route a case, but it never rejects a candidate or ends an employment relationship on its own.
- The quality of the data feeding all of it.
- A workforce plan for your own function.
- New York City: the state comptroller's audit found Local Law 144 had barely been enforced, the kind of finding that comes right before a tougher enforcement phase.
- Illinois's AI employment-disclosure law took effect.
- The EU AI Act's high-risk obligations for employment systems land: risk management, documentation, human oversight and transparency to candidates, with penalties that reach the tens of millions or a slice of global turnover.
- Colorado's revised AI law follows.
In the CHRO's box: ownedDotted line: assigned
The regulators have made this concrete, and the dates aren't comfortably far off anymore. A compliance posture built on "nobody is looking" just expired.
The vendors will keep telling you the AI is ready, and for them it is. Whether it's ready for your people depends on whether your processes are trusted, your data is clean and your governance is real. Those questions belong to the CHRO.
Own them, and AI becomes the lever that finally moves HR from administration to judgment. Skip them, and you've bought the fastest, most confident way yet to scale the things your people already didn't trust.
Sources. SHRM, "State of AI in HR 2026," for 39% HR adoption. Gartner, "CHROs' Top Priorities for 2026" (Oct 2025), for 88% no significant value yet, about 29% of gains from operating-model change, and 82% planning agentic AI. Mercer, "2025/2026 Skills Snapshot," for fewer than one in five skills-based at scale. Deloitte, "Skills-Based Organization" and "2026 Global Human Capital Trends" (66% say functions must change; 7% making progress). Josh Bersin Company, "2026 Imperatives." Yale CELI via Fortune (Apr 2026); World Economic Forum, "Future of Jobs 2025." ServiceNow customer results (Robinhood; City of Raleigh). Regulation: EU AI Act (Aug 2026); Illinois HB 3773 (Jan 2026); Colorado AI Act (Jan 2027); NYC Local Law 144 and the NY State Comptroller audit (Dec 2025). Amazon recruiting-AI bias, widely reported.
Decide what AI touches before the vendor decides for you.
If AI is moving into your people decisions, an independent read on the operating model, the data and the governance is the cheapest insurance you'll buy.