PwC analyzed over a billion job ads across six continents and found AI is creating a two-track labour market. The divide isn't human versus machine. It's the people using AI to multiply their output versus the people being replaced by the smaller team of those who do.
I keep thinking about two people I know - same company, same role, same manager, same AI deployment rollout six months ago. Today they are on completely different career trajectories.
One of them treated the new tools as optional. Not out of stubbornness - he's sharp, he's good at his job, and he genuinely didn't see the urgency. The other one dug in immediately. She started using AI to draft, to synthesize, to prep for meetings faster than she used to prep for the meeting before the meeting. By month three, her output looked like two people's work.
The first person is now being quietly reassessed. Not because AI replaced him. Because the team learned it didn't need as many people doing what he does - when one person with AI could do it better.
That's the actual split.
It isn't humans versus machines. It's the professionals who adapted versus the ones who will be replaced by the smaller team of professionals who did.
What the Data on AI and the Labour Market Actually Shows
PwC's 2026 Global AI Jobs Barometer - which analyzed over a billion job ads across six continents - found that AI is creating a two-track labour market. The finding is blunter than most organizations are ready to hear: the divergence isn't gradual, and it isn't evenly distributed. It's happening function by function, and in some cases person by person within the same team.
What makes this harder to act on is a number from McKinsey's State of AI 2025: 88% of companies are already using AI in at least one function. And 32% of organizations expect headcount reductions in the next year because of it. Those two numbers together tell you something that isn't getting said clearly enough in most leadership conversations. The deployment is already underway. The workforce adjustment is coming behind it.
PwC's own finding sharpens this further: just one in eight CEOs are seeing both increased revenues and reduced costs from AI. Which means most organizations are running AI deployments that are producing outcomes they cannot clearly measure - while the workforce restructuring happens anyway.
Why "We'll Train Everyone" Is Not a Strategy
When I've been working through team structure questions in my current operating environment - figuring out where AI genuinely augments what my team does versus where it changes what we need - the hardest conversations aren't about technology. They're about adoption behavior.
The assumption embedded in most AI rollout plans is that access equals adoption. You give people the tools, you run a few training sessions, and the organization moves forward together. That's not what I've seen. What I've seen is that a subset of people genuinely integrate the tools into how they work, and a larger subset waits - for more training, for clearer use cases, for permission, for some signal that this is the real thing and not the last wave of tools that came and went.
I used to think the waiters would eventually come along as the tools matured. I'm less sure of that now. The gap between the early adopters and the holdouts isn't closing - it's widening as the tools improve and the adopters get faster.
The training sessions don't close that gap. Adoption behavior does.
What Does "AI Adoption Behavior" Actually Look Like at Work?
This is where most articles go abstract. Let me make it specific.
When we started looking at how AI was changing the workflow in our digital marketing operations, the clearest signal wasn't productivity metrics. It was how people handled the beginning of a task. Someone who had integrated AI into their workflow would start a content brief, a competitive analysis, or a campaign audit differently - faster to a workable first draft, more time spent on the thinking rather than the production. Someone who hadn't would spend the same amount of time they always had, doing the same sequence of steps.
The output quality gap started small. After a few months, it wasn't small.
A candidate I interviewed earlier this year said something that stuck: "I don't think of the AI as doing my work - I think of it as the thing that clears the runway so I can actually do my work." That framing landed. It's the right mental model, and the people who have it are not evenly distributed across organizations.
How AI Is Actually Changing Hiring Decisions in 2026
The candidates who had genuinely integrated AI into their workflow stood out immediately - not because they talked about it, but because of how they described their process. They had a different relationship with time and output. They weren't doing the same tasks faster; they were doing different tasks because the production layer had changed.
The ones who hadn't integrated AI weren't bad candidates. Some of them were excellent. But they were pricing themselves at a rate that assumed their production capacity was based on human-only hours. That calculation is changing faster than most people realize.
I'm not making a prediction about who "wins." But I am watching how quickly the cost-per-unit-of-output math changes when one person with AI tools produces what used to require two people without them.
That math is what's creating the two-track labour market PwC identified. Not layoff decisions made in boardrooms. Individual hiring and resourcing calculations made at the team level, month by month.
Which Roles Are Most Exposed to AI Workforce Displacement in 2026?
The PwC data doesn't single out specific job titles, and I'd be cautious about any list that claims precision here. But the pattern in what I'm observing points toward functions where the core work is production rather than judgment - first drafts, synthesis, research compilation, reporting, analysis that feeds a decision rather than being the decision itself.
The people most exposed are not the ones in the lowest-skilled roles. They're often in the mid-tier professional roles where the work is high-volume, follows recognizable patterns, and was previously the domain where you paid for reliable execution. AI handles reliable execution very well.
Where AI is compressing team size most quickly:
- Research and analysis roles where the output is a synthesis, not a recommendation
- Content production where quality is measurable and iteration is fast
- Reporting functions that aggregate data for someone else's decision
Where human judgment is still load-bearing:
- Client relationships where trust is built over time and context is tacit
- Strategy decisions that require organizational knowledge, not just information
- Roles where the work is managing ambiguity rather than resolving it
The honest answer is that most roles have elements of both. The question worth sitting with is which elements dominate - and whether the production layer is shrinking faster than the judgment layer is growing.
What Leaders Should Actually Be Doing About the Workforce Split
The instinct in most organizations is to wait for clarity. Which tools will last, which capabilities will stabilize, which roles will be redefined versus eliminated. That instinct made sense eighteen months ago. It's now a liability.
The organizations that are navigating this well are not the ones with the most sophisticated AI strategy. They're the ones that have named the behavior change they're asking for and are actively distinguishing between the people who are adopting and the people who are waiting.
That sounds simple. In practice, it requires a harder conversation than most leadership teams are having - because the people who are waiting are often performing fine by conventional metrics. They're not underperforming. They're just not building the capacity that the next round of role decisions will be made against.
I don't think there's a clean answer for how to handle that transition fairly. The organizations that figure this out will not get it right for everyone. But the ones that don't name it at all will find themselves making structural decisions reactively, under pressure, and in ways that feel worse for everyone involved.
That's the part I keep coming back to.
