Two-thirds of leaders say designing how people and AI work together matters. 6% say they are leading the way on it.
That stat comes from Deloitte’s 2026 Global Human Capital Trends research, which surveyed more than 9,000 business and HR leaders across 89 countries, and it’s a good look at the story of enterprise AI right now. The tools arrived and the licenses got handed out. The training got scheduled. And then, mostly, the work stayed the same, with AI bolted onto the side of it.
Same tool, different job
In their report, Deloitte tells a small story that explains the gap above. A European telecom added an AI assistant to its customer service team and changed nothing else about the job. Productivity rose five percent. Then the company put 90% of its full rollout budget into redesigning how people and the AI worked together, with new workflows, trust thresholds, escalation paths, and training. Productivity rose thirty percent. Same tool. Different job.
The pattern holds beyond one company. Deloitte found that organizations are twice as likely to exceed their ROI expectations for AI when they prioritize work design, and that the leaders in intentional human-AI design are nearly 2.5 times more likely to report better financial results.
Buying the model or the tool is easy. Redrawing the job around it is the actual work, and at this point, it is nobody’s actual responsibility.
What happens to the other forty percent
Take a recruiter. The role today looks like: source, screen, schedule, interview, synthesize, follow up. AI can now take large parts of the sourcing, the screening, the scheduling, and the synthesis. The tempting response is to call the recruiter forty percent more efficient and move on. The real question is what the other forty percent of their time becomes.
We believe it looks like more time building relationships with candidates. Better calibration with hiring managers. Sharper talent intelligence. Each of those is a choice that can’t be made by the software you use.
At 7-Eleven, a customer can now scan a QR code in the store, text through a short application with an AI assistant named Rita, and have an interview booked before they finish their Slurpee. The old process was slow enough that candidates were often hired somewhere else before anyone called them back. Once Rita took over screening and scheduling, the recruiting coordinator job lost most of its work. 7-Eleven’s head of talent acquisition wrote that the role “simply wasn’t needed anymore. But the people were.” The coordinators moved to the field recruitment team, where they now advise store leaders and help find the right managers for each store.
That is organizational design, and it lands on People Ops.
Everyone agrees but few have started
The data says most companies have not made that choice yet. In a December 2025 Gartner survey of 110 CHROs, 78% agreed that roles and workflows will have to change to get real value from AI, yet just over half had actually redesigned or redefined any roles.
The 2026 CHRO Survey from the CHRO Association shows the same gap from another angle. 47% of CHROs said they had not established clear measures of AI’s effect on productivity, and 48% said redesigning performance management and career progression was on the roadmap but not yet implemented. The biggest barriers they reported were organizational rather than technological, with employee fear of job loss at the top of the list.
Where AI adoption actually stalls
When AI adoption stalls, most companies book another training session. Gallup’s research keeps tracing the difference back to the person’s own manager. Employees whose managers actively back their team’s AI use are 8.7 times more likely to say it has changed how their organization works, and 7.4 times more likely to say it helps them do their best work every day. Picture two recruiters on the same team with the same tool. One has a manager who asks in their weekly one-on-one which tasks she handed off to AI and what she did with the time. The other has a manager who never brings it up. Gallup’s numbers suggest the first recruiter is far more likely to actually change how she works. As of May 2026, only 36% of employees in organizations rolling out AI strongly agreed they had the first kind of manager.
The managers being asked to carry this are stretched thin. Manager engagement fell nine points between 2022 and 2025, from 31% to 22%. And HR leaders know it: half of the CHROs Gallup surveyed said they were not confident in their managers’ ability to guide staff through AI use.
The questions your employees are actually asking have little to do with prompt training. They want to know whether they are allowed to use AI, if they are expected to, if finishing a three-hour task in thirty minutes just earns them more work, and whether their organization is hoping AI makes their role disappear. Those are management questions, and no amount of AI skills training can answer them for you.
The math on automating your job
Under the manager problem sits an incentive problem. Suppose someone finds a way to do seventy percent of their job with AI. If they say so, they may simply be handed more of it. If the role turns out to take a fraction of the time, the role may not survive. So the rational move, in a lot of companies, is to stay quiet and keep the hours.
Most organizations have left that question unanswered. A Gartner survey of 114 HR leaders found that only 7% of organizations give employees guidelines on how to use the time AI saves them. Firms ask people to reinvent how they work without answering the first thing the person wants to know, which is what happens to them if they do.
The new literacy is judgment
Picture a recruiter on a Monday morning with forty AI-generated candidate summaries in her queue. A year or two ago, the skill companies trained for was getting a clean summary out of the tool in the first place. The skill that matters for them today is catching that three of those summaries smoothed over a career gap she would have asked about, and realizing that for the two finalists, reading the resumes herself would have been faster than double-checking the output. Most AI training still teaches the first skill. The second is judgment, and it decides whether the time AI saves is real.
Anthropic’s AI Fluency framework, developed with professors Rick Dakan and Joseph Feller, calls this competency discernment: evaluating the quality of what AI produces and how it arrived there. Discernment carries a hidden cost that rarely shows up in productivity math. Call it the discernment tax, the effort of evaluating and correcting AI output, which can run higher than the task itself. The valuable person is not the one who uses AI the most. It is the one who knows when not to.
Very few companies measure this yet. Only 7% of CHROs in the CHRO Association survey said they were piloting performance metrics focused on judgment, collaboration, and strategic thinking rather than task completion.
People Ops has been handed a different job
Put it all together and we see that People Ops has been handed a different job than the one it was built for. The old question was “what people do we need?” The new one is much more complicated. What work has to happen, which parts should humans do, which parts should AI do, how should the two hand off, and what kind of organization should exist around that arrangement.
This is the judgment premium in practice. As the tasks get cheaper, the judgment left over gets more valuable, and the function that spent decades sizing the workforce now has to design the work itself.
Who teaches judgment now?
There is a harder question underneath all of this, and People Ops is the function that will have to answer it. Judgment is never really taught; it’s accumulated from years of doing the small, unglamorous work. The resumes screened clumsily at first, the schedules that fell apart, the hiring calls that went nowhere and taught you what a good one sounds like. The junior work was the apprenticeship. And the junior work is exactly what AI takes first.
That is already showing up in hiring. Gartner found that 22% of CHROs report at least one business leader in their organization has stopped hiring for entry-level roles because of AI automation, and its analysts warn that cutting early-career pipelines creates serious workforce problems down the road.
So a company can now automate the tasks that used to build judgment skills while calling that same judgment its most valuable asset. Both cannot stay true for long. If the path that produced this generation’s judgment is closing, someone has to build the next one on purpose, because it will not happen on its own anymore.
Where People Ops can start
Recruiting coordination is the easiest place to start, since the tools there are already doing real work at companies like 7-Eleven. Take a coordinator’s week and go through it task by task. Screening questions and interview scheduling can go to the AI. Reference calls and the conversation with a candidate unhappy about an offer stay with a person. Then decide what pulls a task back to a human, like a candidate who replies to the scheduling bot asking whether the role is still remote. The telecom only saw its thirty percent after doing this kind of mapping. That same document is where you answer the question employees are too careful to ask out loud: what happens to the hours they get back.
A recruiter who knows those hours go toward calibrating with hiring managers, and that the work counts toward her next promotion, has a reason to go looking for them. The manager running that team needs protected time on the calendar to lead the change. And the junior coordinator whose scheduling work just disappeared should spend part of the freed time reviewing the AI’s screens alongside a senior recruiter, because that is where the judgment the rest of this post worries about will come from.
The question that People Ops now depends on is: how can a person still provide judgment and taste once the slow, cheap years of getting good have been optimized away? Answer it, and you have built something no model can hand you. Fail to, and you end up with a company full of capable tools and no one left who can tell when they are wrong.
