EY Canada published a piece on May 21, 2026, that deserves more attention than it’s getting.
The article, “AI and people at work: the five tensions that can make — or break — success,” was authored by Darryl Wright, EY Canada’s Work Reimagined Leader, and co-authored by Andy Leung, Director in People Consulting. It draws on EY’s Work Reimagined Survey 2025 and names five specific tensions that, in their words, can derail progress in any organization pursuing AI advantage.
The piece is genuinely useful. Most consulting content of this type is surface-level pattern-matching — “here are five things to think about.” This one is research-backed, counterintuitive, and lands one finding that I think is the most painful single data point in the current AI transformation literature.
I want to do two things in this post. First, walk through what EY actually found, with the research properly cited. Second, build on the framework with what I think they didn’t quite say directly — because the implications are larger than the original article spells out.
The five tensions, as EY frames them
EY’s framework identifies five named tensions every organization pursuing AI advantage must learn to navigate. The most striking one, and the anchor for the entire piece, is the first.
Tension 1: The learning-retention dilemma.
According to EY’s research, the 81-plus hours of AI training that drives adoption excellence also makes employees 55% more likely to leave. Employees with fewer than four hours of AI learning express a 21% intent to quit; that rises to 45% for those with 81 or more hours.
Read that again. The training programs that produce the AI adoption you want are also the programs that cause your best people to leave. Worse, the motivations of highly trained employees shift as they’re trained — employees with more than 40 hours of AI training begin to prioritize “opportunities to work with the latest technology and enhanced flexibility over traditional compensation and career advancement.”
This is the data point that should be reframing how every CHRO in the country is thinking about AI rollouts. I’ll come back to it.
Tension 2: AI-enabled time gains versus rising workload.
AI delivers tangible time savings, with employees reporting an average of eight hours saved per week. However, 64% of employees report an increased individual workload over the past 12 months. While AI is helping save time on existing work, it could also be introducing additional effort.
The time savings aren’t going to rest, recovery, or higher-value work. They’re being absorbed into more work. The leadership question EY puts forward: how do you capitalize on AI time gains without creating an addition rule that erodes the gains?
Tension 3: The anxiety-innovation gap.
While 72% of organizations have found AI has enhanced innovation, 38% of employees fear job loss due to AI. The same 38% worry about overreliance on AI eroding human skills, expertise, and learning. This fear coexists with innovation demands.
Most leaders treat anxiety as an HR problem to be managed downstream of AI rollouts. EY’s framing is sharper: “Prioritize AI anxiety as a leadership capability instead of an HR issue.” That phrase is doing real work.
Tension 4: The shadow AI challenge.
23% to 58% of employees are bringing personal AI tools to work; some are paying for their own subscriptions. Using non-company approved AI introduces risks to safety, security, and oversight.
The range is striking — 23% to 58% depending on the population studied. Even at the low end, more than one in five employees is using AI tools the enterprise hasn’t sanctioned. Most enterprise risk teams have no meaningful visibility into any of it.
Tension 5: Reorganization versus change fatigue.
Roughly 8 out of 10 talent advantage employers have already significantly reorganized due to AI, yet 74% recognize they still need to evolve.
The implication: continuous reorganization is the new operating mode, and the workforce is increasingly running out of capacity to absorb it. EY recommends treating “adaptability as a collective capability” and replacing “the idea of transformation with a philosophy based on continuous adoption.”
The underlying pattern EY identifies but doesn’t fully name
Here’s where I want to build on the framework, because the piece names five tensions but doesn’t quite name the unifying pattern underneath them.
Every one of these tensions is the same underlying problem in different clothes.
The technology is moving faster than the organizational architecture around it.
- The training architecture wasn’t designed for content that doubles in value in six months.
- The compensation architecture wasn’t designed for employees whose market value shifts quarterly.
- The career path architecture wasn’t designed for roles that change shape every year.
- The recognition architecture wasn’t designed for output that AI helps multiply.
- The change management architecture wasn’t designed for continuous, simultaneous, overlapping reorganizations.
For thirty years, enterprise HR systems were optimized for stability. Define the role. Set the comp band. Build the career path. Reward tenure. Promote on performance. The architecture assumed people stayed in roughly the same kind of job for years at a time. The assumption is no longer true.
Highly-trained AI employees aren’t staying in roughly the same kind of job. They’re evolving every six months. Their identity, their value, and their motivations are all changing faster than HR can update the system that’s supposed to retain them.
When EY reports that 81-hour-trained employees prioritize technology access and flexibility over compensation and advancement, they are reporting that the entire incentive architecture of modern enterprise work is misaligned with the people the enterprise most needs to keep.
This is the part the article touches on but doesn’t develop. The fix isn’t to train less. The fix isn’t to throw more money at retention. The fix is to update the entire architecture of how the enterprise hires, develops, rewards, and promotes — to match the new reality of what a highly-trained AI employee actually wants.
Why “train less” is the wrong reading
There’s a reading of the EY data that some executives are going to be tempted toward: we should train less, because training causes turnover.
That reading is going to wreck the organizations that adopt it.
The 81-hour-trained employees aren’t leaving because they were trained. They’re leaving because the organization around them didn’t change to match the person the training created. They became more valuable. The internal job market couldn’t accommodate the new value. So they took the new value to a different employer.
If you train less, you don’t retain people. You just have fewer people capable of leading your AI transformation. You’ve solved a turnover number by destroying the underlying capability.
The right reading is: the training is working. The architecture is broken. Fix the architecture.
What the architecture fix actually looks like
EY’s article lists specific actions for each tension. They’re directionally right. Let me make them sharper.
For the learning-retention dilemma, the question isn’t training volume. It’s career architecture.
Most enterprise career paths still assume the natural advancement direction is up — into people management, larger teams, larger budgets. That path doesn’t match what highly-trained AI employees want. Many of them want lateral movement — into different functions, different projects, different applications of the same technical capability — without losing comp.
Build a parallel technical track that pays as well as the management track and offers explicit AI-adjacent variety. If you don’t, your most valuable people will find that track at a competitor.
For the time-gains versus workload tension, enforce a substitution rule.
EY uses this phrase and it’s the right one. When AI saves an employee 8 hours a week, the default behavior in most organizations is to fill those hours with additional work. The leadership decision is whether those 8 hours go to higher-value work, recovery, or new capability building.
If you don’t make the substitution rule explicit, the default wins. Burnout follows. The AI rollout becomes a productivity story for the company and an exhaustion story for the workforce.
For the anxiety-innovation gap, name it.
The single most powerful thing senior leaders can do here is talk about AI anxiety directly, by name, in meetings, in town halls, in 1:1s. Acknowledge that the people they’re asking to adopt AI also have legitimate fears about what AI means for their careers, their skills, their identity.
Most leaders treat anxiety as an HR matter, downstream of the AI rollout. EY’s framing is correct: anxiety is a leadership capability. The leaders who can name it directly are the ones whose AI programs land. The ones who pretend it isn’t there are the ones whose programs stall.
For shadow AI, replace policing with observability.
EY’s phrase, again. The policing approach (ban personal AI tools, monitor for violations, punish offenders) fails because it pushes shadow AI further into the shadows. The observability approach (provide sanctioned alternatives, monitor what’s actually happening, learn from emergent use patterns) succeeds because it brings the experimentation back into the light.
Some of the most valuable AI use cases in any enterprise are currently happening on personal accounts because the enterprise version is too slow, too restricted, or doesn’t exist. The observability approach captures that signal. The policing approach destroys it.
For change fatigue, the answer is honest acknowledgment.
EY recommends “acknowledging change fatigue as a leadership reality” — which is almost the entire move. Most enterprise leaders treat change fatigue as a complaint to be overcome, not a real organizational constraint to be respected. The data says it’s the latter.
If 80% of organizations have already reorganized and 74% know they need to do it again, you are operating in a workforce that is structurally exhausted. Pretending otherwise produces resentment without producing change.
The deeper implication for Canadian enterprises
There’s one more layer EY’s piece points toward without quite spelling out.
EY’s research shows organizations with more strategic people and tech capabilities are 17x more likely to outperform in current economic conditions. The phrasing is important: it’s not “people OR tech.” It’s “people AND tech.” Combining talent and AI advantages — not one or the other — is what produces outperformance.
This is the most important sentence in the article, and it’s buried near the top.
The Canadian enterprises most likely to win the next decade aren’t the ones with the best AI strategy. They aren’t the ones with the best talent strategy. They are the ones whose AI strategy and talent strategy are designed together, by the same leadership, with the same horizon, with the same incentives.
In most enterprises, these are two separate streams. The CIO owns AI. The CHRO owns talent. They report to the same CEO but operate on different cycles, different budgets, different metrics, different vendors. The integration EY is calling for doesn’t happen because nobody has been organized to make it happen.
The first Canadian enterprises to break that silo — to put AI and talent strategy under a single integrated leadership, with shared metrics and shared horizons — are going to capture an outsized share of the 17x outperformance EY is measuring. The ones that maintain the silo will be the ones generating the turnover data EY is reporting.
The closing thought
EY’s five tensions are a useful lens. The real insight is in what they collectively reveal: every dimension of the modern enterprise employment relationship is straining under the speed of AI change, and the organizational architecture that was built to support it was designed for a world that no longer exists.
The leaders who’ll thrive in the next decade aren’t the ones who’ll out-train, out-spend, or out-reorganize their competitors. They’ll be the ones who rebuild the entire architecture — career paths, comp structures, recognition systems, change cadences, anxiety management — to match what a highly-trained AI workforce actually wants.
The single warning number from the EY data is 45%. That’s the intent-to-quit rate for your most-trained AI employees. The data is telling you, with unusual specificity, that the people you most need to keep are the people most likely to leave.
This week, ask one question of your CHRO:
“Have we updated our career paths, comp structure, and reward systems to account for what a highly-trained AI employee actually wants?”
If the answer is no — and in most Canadian enterprises right now, it is — that’s the most important gap in your AI program. Not the technology. Not the platforms. Not the vendor evaluations.
The architecture of how you treat the people you trained.
EY just gave you the warning number. The question is whether you’ll act on it before your 45% becomes your turnover report.
The world has changed. The leaders who notice will be the ones the next decade is built around.
