What HR Teams Should Do Differently In An AI-Driven Work Era
There is a question a lot of people are quietly asking themselves right now, at every level of an organization: am I automating my way out of a job?
HR leaders are asking it too, and that is worth saying out loud, because people teams are carrying something particular. We are being asked to design the organization’s entire approach to AI — the framework, the guardrails, what gets shared and what does not — while upskilling ourselves at the same time, in roles that are changing underneath us as we work.
And the data suggests most organizations have not yet caught up with their own change. A survey of more than 1,000 HR and people leaders conducted for Oyster’s 2026 Global AI Talent Report found 95% of HR and people leaders said employees use AI tools daily in roles where AI was never part of the job description. Among companies that added AI requirements to existing roles in the past year, 62% changed neither the title nor the pay. Among those extending AI expectations into non-technical work, 46% never formally told employees the expectation had changed.
So what should people teams do about it?
Start by accepting that not everyone needs to be a builder
There is a real difference between someone who builds automated workflows and someone who knows enough to apply AI well inside their own role while also contributing to broader organizational goals. Most organizations need far more of the second than the first. AI has a place in nearly every role, but that does not mean every role becomes a technical one.
Within my own team, we’re distinguishing between roles that require deeper expertise in building automations and those focused on using AI and automation to meaningfully improve individual and team performance.
Getting that distinction right is critical. Expect everyone to become a builder and you’ll judge much of your organization against a standard their roles don’t require. Set expectations too low and nothing changes.
Catch more HRTech Insights: HRTech Interview with Dave Hilton, CPO at O.C. Tanner
Use a sharper test than efficiency
Before asking whether a process can be automated, ask whether it needs to exist at all. If it does, consider how you would design it from scratch with AI available today. The goal is not to make a broken process run faster, but to build one that serves its purpose.
Then ask where human judgment or support matters. Automate the moments that do not need a person, so there is someone available for the ones that do.
That is easier to see in practice than in principle. Someone asking a routine policy question — what statutory maternity leave looks like in a particular market — should get that answer quickly, without waiting on anyone. Someone navigating a complicated or sensitive situation involving their own leave should be talking to a human. Those are different moments, and the work is in knowing which is which.
The same logic applies further out. A role with “relationships” in the title should not spend most of its time assembling data to work out how those relationships are going. HR business partners should be partnering, not generating documents. In each case the AI is not replacing the role, it is clearing what was preventing the role from being done properly.
Answer the compensation question honestly
Sooner or later someone will ask whether AI capability should change what a role pays. The typical HR response is: it depends. True, but not useful on its own. So here is the test underneath it. Is the role fundamentally shifting, or are we using a tool that makes the existing work a bit different?
Often it is the second. You remove some work and replace it with higher-impact work, the work people were meant to be doing all along, but never had the capacity to reach. The gap was rarely capability. It was capacity.
That reasoning will not hold for every role, and organizations will land in different places. What matters is that the decision gets made deliberately and communicated, rather than left ambiguous. Ambiguity is where people draw their own conclusions about what the company values, and they rarely guess generously.
The piece nobody is discussing: leveling
Go back to that distinction between builders and people whose roles are becoming more technical, and a harder question emerges: what do those new technical expectations mean for job levels?
The answer affects both the level someone enters the business at and how they progress. We need to define which roles require automation-building expertise, which require other AI skills, and what proficiency looks like at each level. Those expectations need to be reflected clearly in our competencies and leveling frameworks so they are applied consistently across teams. Leave each team to work it out independently without calibration, role by role, and those decisions can become a structural problem. I expect many organizations to revisit their competencies, frameworks, and leveling within a year or two. Better to be deliberate about it now than to inherit the gaps later.
Change your feedback cadence as expectations change
The annual performance review was already a weak instrument. On its own, it cannot keep pace with roles shifting this quickly. When expectations are changing, regular, ongoing feedback stops being good practice and becomes a requirement for setting people up to succeed.
If AI competencies are making their way into your framework, timing matters. You can’t communicate new expectations in December and assess people against them in January. People need time to understand what is expected, training and support to build their skills, and opportunities to practice. Those conversations need to start well before the review cycle.
This gets especially messy where performance reviews are tied directly to compensation. Non-technical roles that now carry a technical element come with a genuine learning curve. Rating people against a bar they were never told about, and never given time to clear, isn’t a performance process. It’s a trap.
Make the time explicit
The most common thing I hear from other people leaders is a version of the same question: where do we start, and how do we make the space to do this at all? You need time to build the thing that eventually gives you time back, and in the meantime the work does not stop. That cycle does not resolve on its own.
The organizations getting this right are treating upskilling as part of the work rather than something stacked on top of it, and they are giving people genuine space and resources to get there. The most effective version I have seen is internal rather than outsourced. People at different stages of fluency learning from each other, applied to problems the organization already has. The hard part is never the tool. It is the organizational context the tool has to fit into.
What this is really about
Twenty-one percent of HR leaders from the same survey said an AI tool or self-service system had already missed something important in a high-stakes employee situation — a termination, a leave request, the moments that have real consequences for a real person.
That number is the argument for everything above. We should absolutely get rid of the monotonous and the tedious. But human judgment has to stay in the moments that matter, and knowing where that line sits is now a core people capability rather than a nice-to-have.
Automate the moments that do not need a person. Keep a human in the moments that do. Everything else — the competencies, the leveling, the comp decisions, the review cadence — is the work of making that real.
About Oyster
Oyster is a global employment platform that helps companies hire the best people anywhere and treat them right everywhere. As the only B Corp-certified Employer of Record, Oyster gives companies both modern technology and in-house expert support so they can navigate international hiring with confidence, clarity, and speed.
Read More on Hrtech : Tech Hiring Is Cooling, Not Stalling. Here’s What Employers Should Prioritize Now
[To share your insights with us, please write to psen@itechseries.com ]