A
measurable divide is opening in the global workforce, and the line runs not
between industries but within them. Workers in the same organisation,
performing related functions, are experiencing opposite trajectories
depending on one variable: whether their role has been augmented by AI tools
or targeted for automation by them.
The World Economic Forum’s Future of Jobs Report 2025 found that
170 million new roles were expected to be created by AI-related demand
through 2030, while 92 million existing positions were expected to be
displaced, producing a net gain of 78 million jobs globally. That headline
figure is frequently cited to argue that AI creates more work than it
eliminates. What it conceals is more instructive: the new roles and the
displaced roles are not in the same sectors, do not require the same skills
and are not accessible to the same workers. The workplace divide is not
primarily between people who have jobs and people who do not. It is between
people whose skills are becoming more valuable and people whose skills are
being repriced downward faster than they can adapt.
Where the Divide Is Already Visible
The clearest current evidence comes from knowledge work, where the
impact of generative AI is neither uniformly positive nor uniformly negative
but strongly differentiated by task type. Research from MIT
Sloan’s Institute for Work and Employment Research found that
generative AI tools consistently boosted productivity for high-skilled
workers performing complex analytical and communicative tasks, while having
neutral or negative effects on mid-skilled workers in roles centred on
routine information processing. The mechanism is straightforward: AI
complements skills that involve judgement, context and synthesis, and
competes with skills that involve retrieval, formatting and routine
transformation of structured information.
For workers in the first category, the tools function as a
productivity multiplier that makes their existing expertise more valuable,
faster to apply and more legible to employers. For workers in the second
category, the tools perform the function they were being paid to perform, at
lower cost and higher speed than they can match. The relative position of
these two groups, already unequal before AI became a workplace reality, is
widening in sectors where deployment has moved beyond experimentation into
operational integration.
The Geography of the Divide
The workplace divide has a geography that public discussion has
substantially understated. Within wealthy economies, the divide maps onto
educational attainment and urban versus non-urban labour markets. Across
economies, it maps onto the existing distribution of digital infrastructure,
AI investment and regulatory frameworks. Countries that have invested heavily
in digital infrastructure, technical education and adaptable labour market
policies are seeing AI generate measurable productivity gains and new
high-value role categories. Countries that have not are seeing automation of
existing work without the compensating creation of new work, because the
investment conditions and skill supply required to generate AI-adjacent roles
do not exist at the necessary scale.
The WEF
Future of Jobs Report identified a growing skills mismatch as one
of the primary drivers of this geographic and demographic polarisation. The
skills most in demand by employers deploying AI, which include analytical
reasoning, complex problem-solving, creative synthesis and the interpersonal
capabilities required to work alongside and oversee AI systems, are not the
skills concentrated in the workers most exposed to displacement risk. Closing
that mismatch requires investment in education and reskilling at a scale and
speed that most institutions and governments are not currently
delivering.
What Changes Inside Organisations
Within individual organisations, the AI workplace divide is
reshaping reporting structures, performance expectations and compensation
decisions in ways that are not yet fully visible in aggregate labour market
data. Roles that were previously defined by the volume of information they
processed are being redefined around the quality of judgement applied to
AI-generated outputs. That redefinition benefits workers who have developed
strong contextual and evaluative skills and disfavours workers who were
compensated primarily for speed and volume of throughput.
The organisations navigating this transition most effectively are
those that have been deliberate about three things: identifying which roles
are being augmented versus which are being automated, investing in reskilling
for the workers in the second category before the displacement becomes
terminal and restructuring compensation to reflect the new value distribution
rather than preserving legacy pay structures that no longer map to actual
contribution. Most organisations have not yet been this deliberate. The
pressure to realise efficiency gains quickly is creating a pattern of fast
deployment and slow reskilling that concentrates the benefits of AI in
existing high-value roles and concentrates the costs in already-vulnerable positions.
For workers trying to understand where they sit in this reshaping,
our analysis of what
the 2025 and 2026 employment data actually shows about AI and job
displacement examines the evidence more precisely than the broad
figures usually cited in public debate. The pattern that emerges is one of
accelerating divergence rather than uniform disruption: some roles becoming
dramatically more productive and valuable, others becoming progressively less
so, at a pace that makes the reskilling window shorter than most workers or
their employers have assumed.
The Skills That Will Matter
The most durable finding from the employment research literature
on AI and work is that the skills most resistant to automation are the ones
that require situational judgement, emotional intelligence, creative
synthesis and the ability to navigate novel problems that do not fit patterns
from past data. These are not skills that most educational systems teach
explicitly or that most employers have invested heavily in developing. They
are skills that workers typically develop through sustained exposure to
complex, ambiguous work, which is precisely the kind of work that AI is being
deployed to assist with rather than replace.
The practical implication is counterintuitive: the most effective
preparation for an AI-saturated workplace is not learning to operate AI
tools, though that is useful, but developing the contextual and evaluative
capabilities that AI tools require human oversight to perform. Workers who
understand the domain their AI tools are operating in well enough to catch
errors, assess outputs and make judgements that the tool cannot make are
positioned very differently from workers who are proficient at using the
tools but cannot independently evaluate what they produce.
The organisations and educational systems that understand this
distinction will produce workers who are positioned to benefit from AI
augmentation. Those that treat AI literacy primarily as tool proficiency will
produce workers who are efficiently replaceable by the tools they have been
taught to use. Our piece on how
AI is simultaneously helping and hurting workers in ways that depend on how
it is deployed examines the specific conditions under which
augmentation and automation diverge in practice. The workplace divide is not
predetermined. It is the outcome of decisions being made now by employers,
policymakers and workers about which direction to push it. For a sharper view
of what that data shows at the individual and household level, our piece on
what
genuine AI bias guardrails require in practice covers the
institutional obligations that make AI deployment equitable rather than
simply efficient.
What Comes Next
The workplace divide created by AI will not close automatically as
the technology matures. The evidence from earlier waves of automation
suggests the opposite: early divergence tends to become entrenched as capital
flows to the roles and sectors where productivity gains are realised and away
from the roles and sectors where displacement is occurring. Without
deliberate policy and organisational intervention, the AI workplace divide
will harden into a structural feature of labour markets in ways that prove
very difficult to reverse.
The interventions that have historically narrowed
technology-driven workplace divides share common features: they invest in
skills before displacement becomes irreversible rather than after, they
distribute the productivity gains from technology broadly enough to sustain
consumer demand and political legitimacy, and they adjust the institutional
rules governing work fast enough to reflect new value distributions rather
than preserving legacy arrangements that no longer match economic reality.
All three conditions are currently absent or partial in most major economies.
That is the most honest description of where the AI workplace divide stands:
visible, widening and not yet addressed with the seriousness its trajectory
warrants.
About the Author
By Stuart Kerr, Technology Correspondent, LiveAIWire. Stuart
covers artificial intelligence and emerging technology, with a focus on how
these developments reshape work, creative industries and everyday
life.