‘It basically underlies everything we do in HR’

Most employers expect their people to be using AI at work. Far fewer have written that expectation down — and that could be problematic.

Ninety-five per cent of HR professionals say employees at their company use AI tools daily in roles where AI is not part of the official job description, according to a report.

Two-thirds have added AI skill requirements to traditionally non-technical roles such as marketing, HR and operations — but of those, 46 per cent say the change was never formally communicated to employees.

The share of roles explicitly requiring AI skills tripled from about five per cent in 2022 to 15 per cent in the first half of 2026, says the 2026 Global AI Talent Report from global employment platform Oyster.

The report draws on a survey of more than 1,000 HR and people leaders at manager level or above, in August 2026, across Canada, the United States, the United Kingdom and Europe, along with an analysis of more than 15,000 job descriptions for roles filled through Oyster.

“Some organizations are afraid to, I think, communicate in a fully honest, transparent way with their workforce. And almost always when we see that, it’s a bad idea on the part of organizations,” says Frederick Morgeson, Eli Broad professor of management at Michigan State University.

“[You’re] better off clearly and transparently communicating what’s going on. And if you don’t know as a leader, you can say, ‘We’re not really sure.’”

‘It basically underlies everything we do in HR’

The gap between what people do and what their job descriptions say isn’t a simple paperwork problem, he says.

“I think it’s super important to have accurate, updated job descriptions because job descriptions and an understanding of what is in the job basically underlies everything we do in HR, whether we realize it or not,” he says.

Hiring flows from knowing what knowledge, skills and abilities a role requires, he says, and so does nearly everything else.

“If you’re going to design a performance management system, you need to know what people are doing and what those important parts of the job are.”

The same goes for pay, in knowing the basis for compensation, or development, in knowing how you’re developing people, says Morgeson.

“If the job is changing, then that just puts pressure on HR to make sure those descriptions are up to date.”

He’s not persuaded by the argument that jobs are “dynamic” and change too fast to be pinned down.

“Even in jobs [where] technology or other changes have tremendous influence, there’s some week-to-week, month-to-month, year-to-year stability. And we should have a sense of what those things are.”

That means doing the work of understanding what is actually changing.

“The question I would have is… ‘What is AI taking on? What is AI taking away from either entire jobs or parts of people’s jobs?’ The jobs get reconfigured. Understanding all that, it becomes important.”

How bloated descriptions hide good candidates

Joseph Fuller, professor of management practice at Harvard Business School and co-lead of its Managing the Future of Work project says the problem predates AI.

“Companies are very lazy about updating job descriptions. And when they do update them, what they tend to do is add material, as opposed to edit.”

That bloat has a direct cost in hiring, says Fuller, because of AI already sitting in the recruiting process. Applicant tracking systems use natural language processing to compare an application against the job description, looking for overlap and keywords.

 “The more long and detailed these things get, the more they have the effect of potentially excluding candidates who actually have the most relevant skills to succeed in the job.”

The result, he says, is that companies gravitate toward people who already hold the same job — or the adjacent one — somewhere else. Fuller has described the people screened out as “hidden workers”: not because they’re hiding from work, but because “employers’ view of them is essentially blocked by the way the AI operates.”

Like email and spreadsheets — only bigger

These kind of transitions have happened before, says Morgeson pointing to spreadsheets replacing hand calculation, and word processors, “just not at the scale and not at the level of sophistication that seems to be happening now,” he says. “That’s where the really interesting challenge comes in for organizations.”

Morgeson says the challenge for HR is the tools have become cheaper and easier to use, but it has become “much more difficult now to understand all the different use cases, all the different places that you can use it.”

That ambiguity shows up in the Oyster data too, he says. “They talk about AI skills, but never really describe what an AI skill is.”

Rewriting competencies

Erin Goodey, director of people services at Oyster, draws the same comparison.

“If we think back to when email first came into the workplace, or when we’re learning new tech, new tools, I think there’s always a learning curve,” she says. “I think [AI] is like the fundamental shift in the way we work, though.”

At Oyster, Goodey says the company is rewriting its internal competencies to make that distinction. Not every role needs deep technical expertise, she says — building automations in a tool like n8n is very different from using ChatGPT, Claude or Notion AI.

An “automation builder” expectation is being assigned to specialized roles, while the company-wide bar is using AI to transform your own work.

“Not everybody needs to be like a builder,” she says. “But… there’s a place for AI in probably every role.”

Don’t save it for the performance review

According to the report, new AI requirements usually arrive through a manager rather than a formal process. Fuller says his own research found a similar disconnect.

“We did some interesting research that said that 90% of supervisors expected their subordinates to use AI. But, at the same time, 80% of workers said, ‘I wish my supervisor would tell me what it is they expect me to do with AI,’” he says.

Could the performance review be the place to set those expectations? Morgeson sees a problem with relying on that approach if people were never told about these expectations.

The review may be where accountability gets applied, he says, but “if you’re going to hold somebody accountable to something, that should be at minimum communicated to them and maybe more formally incorporated somehow into a job description or some other documentation.”

Part of the difficulty is that managers themselves are still figuring it out, says Morgeson.

“It does feel a little bit like it’s the blind leading the blind in a way. So, I’m a manager. I’m not exactly sure how this thing’s impacting us, but I know it’s important.”

‘Messy and dangerous’

Goodey argues the annual cycle is too slow for this.

“If you’re having an annual performance appraisal, you shouldn’t be kicking that off in January and then telling everyone in December what your competencies are that you’re going to be rated against. And if AI is finding its way into that competency framework, then you need to start having those conversations around expectations much earlier.”

She also warns about the link between reviews and pay.

“When you are in an organization where your performance appraisals are actually also tied to your comp cycle, I think that’s when it’s going to get a little messy and dangerous, and you need to figure out how you decouple those things.”

Should pay change? ‘AI can go both ways’

Among companies that added AI requirements to existing roles in the past 12 months, 62 per cent didn’t change either the title or the compensation, found Oyster. And among the roughly two-thirds of companies that ask candidates what they’ve personally built with AI, 53 per cent say the answer doesn’t affect hiring or pay decisions.

Morgeson says the pay factor depends on what AI is doing to the job. Pay is typically set either through compensable factors — the level of skill and decision-making a job requires — or by what the market will pay for scarce skills.

“I think the question… on the pay side is ‘Is AI making the work more complicated, making the work require higher-level skills that are reflected in skills that are hard to get in the labour market? Or is it making a job simpler?’ Because to me, it feels like AI can go both ways,” he says.

If an AI-based system takes over the complex decision-making and a role is reduced to taking customer orders, those jobs are simpler, he says: “So, yeah, I think in that case the pay would go down.”

But if workers now must run the prompts, interpret the output and verify its accuracy, “that seems like that definitely makes the job more complicated,” says Morgeson.

Productivity and pay

Fuller doesn’t see a blanket case for raises.

“I don’t think there’s an umbrella answer to that. I think in a lot of jobs, this is just a new way to be productive,” he says.

“In some ways, it’s a problem that takes care of itself, because if you’ve really developed deep skills in AI such that you’re highly more productive, that’s going to qualify you for jobs elsewhere or jobs inside your organization that pay more.”

Goodey says her response is typical for HR: “It depends.”

“Is the role fundamentally shifting? Is the role actually quite different, or are we leveraging a tool that’s making the work a little bit different?”

At Oyster, she says, it hasn’t generally been a compensation discussion.

“Yes, we’re getting rid of some work, but we’re replacing that with other work. And that other work is actually more of the high-impact, high-value work that likely those folks should have been doing anyway, but the gap was actually the capacity.”

She expects companies to revisit competencies, frameworks and job levelling within a year or two.

“If not everyone is a builder, but everyone has this new tech skill requirement, what does that mean in terms of the level in which they come into the business or the levels in which they can progress through the business as well?”

Why some workers keep AI use quiet

When an employee automates part of their own job, only 38 per cent of HR professionals in the Oyster survey say that person would be rewarded. Twenty-six per cent say nothing would change, 24 per cent say they’d be given more work and six per cent say the role would shrink.

Fuller says companies generally don’t plan to pay people more for using AI, and that if it noticeably cuts someone’s workload, “they’ll find some more things for them to do.”

“You don’t have a positive incentive as a worker to point out that you’re making extensive use of it. You arguably have a negative incentive, disincentive to report it,” he says.

Building AI skills in-house

When AI skills are missing, 53 per cent of companies in the survey prefer to train existing employees, 31 per cent combine training with hiring and 14 per cent primarily hire externally, found Oyster's report.

Training is likely the fastest, easiest route given the time and cost of hiring, says Morgeson — and it sends a message. “That data point suggests that, at least at some level, organizations care enough to try to upskill people rather than just replace them.”

Goodey says internal learning teams can tie AI to “how we do what we do and why we do what we do.” Oyster has run AI accelerator cohorts, pairing employees at different stages with coaches and more technical colleagues who help them turn a real company problem into an AI solution.

“I think you get that so much more organically and naturally when it’s internal versus an external provider coming in and trying to figure out all the nuances and the hows behind the problem that you’re trying to solve,” she says.

But employers have to make room for it.

“Upskilling is part of the work rather than something that’s on top of the work,” she says. “We need to, as organizations, create the time and space for people to be able to develop the skills.”