Effective AI training starts with experimentation, not tools, experts say
While artificial intelligence (AI) adoption is accelerating across workplaces, access to the training and resources meant to support it is falling — and for many workers, that gap is starting to show, according to a new global report.
AI use among workers rose 10 percentage points year over year, with nearly two-thirds (64 per cent) of the workforce now using it at work.
But access to training and development fell eight points, with 51 per cent of workers reporting access to the resources they need in 2026, down from 59 per cent the year prior, according to PwC’s 2026 Global Workforce Hopes and Fears Survey, which polled 49,364 workers across 48 countries and regions.
For HR leaders, the data raises an important question about what proper training and development looks like in practice.
For Fiona Ho, CHRO at Sophos, a cybersecurity technology company based in Vancouver, the answer starts with access.
“The approach we’ve taken at Sophos is ensuring that everyone first has access to AI so they can experiment,” says Ho. “Once people have experimented, they can think about: what am I doing in my day job that I could do differently, more effectively? ‘Where is there lower-value work I could use AI for, so I can spend my time on higher-value work?”
What does good AI training look like?
At Sophos, the training doesn’t start with AI tools themselves. Ho says the focus is on helping employees think about how those tools can solve problems — or pain points — they already face in their daily work.
“We’ve made what we call AI-enabled curriculums available for employees. We have employee curriculums and manager curriculums, and they effectively complement each other,” she says. “But it’s not about training for the sake of training. We tie it back to how it’s going to impact what we’re doing day to day — finding opportunities, understanding pain points, and helping employees solve real problems.”
Ho points to one training module as an example that asks employees to rethink meetings altogether, using AI summaries, video recaps, and optional attendance to cut time spent in rooms where decisions aren’t being made.
“That’s a simple example of identifying a real pain point — too many meetings, not enough space for heads-down work — and finding ways to solve it for employees,” she says.
Community-based learning
Beyond the curriculum itself, Sophos has also leaned into community-based learning as an ongoing resource, Ho says.
Those groups allow a range of employees to come together to showcase their work and learn from each other and play a key role in helping employees who are falling behind build much-needed confidence, she says.
According to the PwC report, AI adoption is splitting the workforce. It identifies two key groups: “front-runners,” 14 per cent of workers who combine scarce skills with strong AI capabilities, and the “engine room,” the 56 per cent majority whose skills are less scarce and who are further behind on AI.
Fewer than 40 per cent of engine room workers say they have access to the learning and development they need, and only half feel confident about their job security, compared with 85 per cent of front-runners.
Ho notes that expecting every employee to reach the same level of AI literacy is neither practical nor realistic.
“It really comes down to the individual. We can provide the space for conversations with colleagues and leaders, and opportunities to experiment. But there also needs to be drive coming from the employee. It’s really about meeting in the middle,” she says.
Gap between AI use and training
Emmanuelle Vaast, a professor and dean of Desautels Faculty of Management at McGill University in Montreal, says the same pattern shows up in research — and closing the gap starts with recognizing that different employees have fundamentally different needs when it comes to AI.
“An entry-level employee is not going to have the same AI needs as a manager,” she says.
She flags something AI training programs often miss — not every employee who lacks AI skills is simply behind. Some are actively resistant.
What research is beginning to show is that some employees, across all levels and job types, develop what Vasst calls “algorithmic aversion” — a resistance to AI that has less to do with skill and more to do with how people perceive the technology and what it means for their professional identity.
“For these employees, just training them on the technical side is not going to work,” she says. “There need to be discussions, communication, and an accompanying process to show them that the tool can actually help them do a better job.”
If employers really want to close the gap, organizations should focus on helping employees understand the technology and the hands-on experience needed to work with AI outputs, Vasst says. That can include workshops where outside experts explain the foundations and concepts of generative AI.
“That foundation is really important for people to then use it correctly,” she says.
The core reason the resource gap exists, Vasst says, is the speed at which AI moves — which makes it hard for companies to keep up. Employers would need to be continuously testing AI tools to keep their training current.
“For a company to develop training resources and then deliver them to employees, they’re likely to always be six months behind. And six months in AI terms is a significant amount of time,” she says.
The time and manager problem
But even where updated training exists, it is not enough, says Nabil Beitinjaneh, a professor at McGill University and AI specialist.
The more pressing question is whether employees are given the time to learn and apply what they are being taught — which, for the most part, he says they are not.
A 2026 study of 1,000 employees in the United States found that 56 per cent reported having no properly allocated time to learn AI skills, according to Workera’s 2026 State of Skills Intelligence report.
“This is the biggest challenge in any organization. With acceleration, demands go up, expectations go up — yet people are not necessarily given the time and tools to keep pace,” he says. “You can watch how-to-ride-a-bicycle videos forever — that doesn’t mean you know how to ride a bicycle. You have to go out and do it.”
That time gap, Beitinjaneh explains, trickles directly down to managers — who he argues are both the most important and most overlooked lever in any AI workforce transformation as they are often being asked to lead change, but haven’t been equipped to lead themselves.
“The people manager is not recognized as such and is not given the time and tools to do their actual work — which is to develop their people, move them forward, and build trust and understanding,” he says.
It is an investment, he argues, that organizations need to start making.
Measuring if training is working
With all the complexity of building the right training programs, the question becomes how to evaluate whether they are actually working.
Ho says the measure should be thought of through output and quality of work.
“When you look at the results, the work, and the productivity — is it better quality? Are the outputs stronger? That would be a way to assess how effective the training is. It all comes down to the performance of individuals,” she says.
