Companies are asking their employees to become proficient with AI. But what does that actually mean?
Last week at Gartner HR Symposium in Orlando, we asked HR, learning, and talent leaders how they were using AI themselves. Not what their organizations had announced or what their technology roadmaps promised. We wanted to know what they were doing with AI day to day, what was working, and where they were running into trouble.
The conversations were revealing.
Some leaders were using AI to prepare presentations, draft communications, and conduct research. Others were hesitant to use it for anything involving employees. Several were learning on their own, without much guidance on what good AI-assisted work should look like.
One tension kept coming up: HR is being asked to help the rest of the organization adopt AI, while much of its own AI adoption is still being figured out.
Here are five things that stood out.
Everyone needs AI skills. Few can define them.
AI proficiency is quickly becoming an expectation of professional work. But it’s not necessarily becoming part of how companies define jobs.
Of the six leaders we asked about role expectations, none pointed to AI skills explicitly documented in their job descriptions or role profiles.
That disconnect extends beyond the people we spoke with. A 2026 Zapier study found that 98% of executives expect employees to have some level of AI proficiency. Meanwhile, Deloitte reports that 89% of life sciences and healthcare organizations haven’t redesigned jobs around AI.
The question is what proficiency actually looks like.
For someone in talent acquisition, it might mean knowing how to use AI for initial research without compromising candidate evaluation. For an HR business partner, it could mean interpreting organizational data, checking the quality of an analysis, or identifying where human judgment is essential.
These are different skills, applied to different decisions. Yet they’re often grouped under the same broad expectation to become AI proficient.
Until companies define what effective AI use looks like in each role, it’s difficult to know what to hire for, what to teach, or how to measure progress.
Training has become a lesson in what not to do.
When we asked six leaders about their AI training, two said they received none. Three described annual or mandatory sessions focused primarily on acceptable use, restrictions, and compliance.
Those rules matter, especially when working with sensitive employee data. But they address only part of the problem.
The most frequently mentioned skills gap was prompting, raised by four people. Judgment and critical thinking followed.
Learning what information can safely go into an AI tool is different from knowing how to use it well.
Employees need to understand how to frame a question, recognize a weak answer, and decide when an output is useful enough to act on.
The broader research suggests this is far from solved. A University of Melbourne and KPMG study found that only 47% of employees globally had received any AI training.
Companies have begun teaching employees how to avoid misusing AI. Fewer seem to have figured out how to teach them to produce excellent work with it.
The first draft is getting faster. The finished work isn’t always.
Presentations were the most common AI use case in our conversations, mentioned by eight of the 20 people we spoke with.
Four described having to rework AI-generated drafts to make them appropriate for their audience, tone, or brand.
This is a familiar pattern. AI makes it remarkably easy to produce something that looks finished. Determining whether it’s actually good still takes work.
BetterUp Labs and Stanford researchers found that 40% of workers had received what they described as AI “workslop” in the preceding month, requiring roughly two hours of rework per instance.
Of course, editing has always been part of professional work. The question is whether AI is reducing the total effort or simply moving it from creation to correction.
A presentation that takes five minutes to draft but an hour to fix isn’t necessarily an improvement.
And when employees are using AI to produce work that others must review, the costs of that rework can fall on someone else.
Measuring the speed of the first draft tells only part of the story. What’s more interesting is how much time it takes to get to a result worth using.
The rules are unclear, so employees are making their own.
The leaders we spoke with described very different approaches to AI governance.
One said restrictions on approved tools were slowing down work. Another described avoiding AI for candidate and performance assessment altogether. A third was already using it to support first-pass performance reviews.
These aren’t necessarily contradictory positions. They reflect how organizations are drawing different boundaries around the same technology.
The difficulty is that employees still need to get their work done, whether or not those boundaries are clear.
Research from the University of Melbourne and KPMG found that only 40% of workplaces had a generative AI policy, while 57% of employees reported hiding their AI use.
That’s a significant visibility problem.
When leaders don’t understand how employees are using AI, they have less information to make good decisions about where to encourage adoption, where to introduce safeguards, and where existing tools are falling short.
Governance needs to do more than restrict certain behaviors. It needs to reflect how work is actually happening.
AI is changing how companies evaluate candidates, too.
One talent acquisition leader raised a particularly difficult problem: knowing whether the person applying for a job is who they appear to be.
They estimated that 20% to 30% of applicants for many roles were either inauthentic or so heavily tailored to the job description that it was difficult to assess their actual qualifications.
That was one leader’s estimate, not a measured rate across the organizations we interviewed. But the concern is increasingly familiar.
Gartner has forecast that by 2028, one in four candidate profiles worldwide could be fake.
For talent teams, this introduces a new challenge to an already complicated process.
AI can help candidates communicate their experience more effectively. It can also make applications harder to distinguish, and give fraudulent applicants new ways to appear credible.
Hiring teams now need to assess not just whether a candidate’s qualifications match a role, but how reliably those qualifications can be established.
The same technology that promises to make recruiting more efficient is creating new work around verification and judgment.
The next challenge isn’t getting people to use AI.
Across these conversations, most leaders weren’t debating whether AI had a place in their work. They were trying to understand how to use it responsibly, consistently, and well.
That distinction matters.
It’s possible for an organization to have high AI adoption and still struggle with output quality, unclear expectations, unnecessary rework, or employees working around its policies.
A dashboard can show how many people are using an AI tool. It won’t necessarily explain whether that tool is improving their work.
For HR leaders, the more useful starting point may be to understand what’s happening inside individual roles. Where is AI genuinely helping? Where are employees spending time correcting its output? What are they avoiding, and why?
Those answers can inform much more specific training, clearer expectations, and better governance.
Before organizations can decide how to accelerate AI adoption, they need a clearer picture of what adoption actually looks like.
That’s a question worth asking the people closest to the work.
About the research: These observations are based on 20 informal, anonymous conversations conducted at Arist’s booth at Gartner HR Symposium/Xpo, October 27–28, 2026. Not every participant answered every question. Counts reflect the specific conversations reported, not a representative survey. Arist is not affiliated with Gartner.
