The conditional buried in the research

The exposure studies that dominate coverage of AI and employment contain a clause that rarely survives summarising. ILO Working Paper 140 sets out what determines whether an exposed occupation actually disappears, and gives two factors: "the initial decision to adopt the technology, but also the extent to which individuals in these occupations are given opportunities to learn to work with these technologies and adapt to the evolving nature of their tasks".

Both factors are employer decisions. Neither is a property of the technology. This is the single most consequential thing in the literature for anyone running an organisation, and it is almost always dropped in favour of the headline percentage.

The ILO adds an economic argument for the same conclusion, citing Acemoglu and Autor: the productivity benefits of the technology lie "not in the search of outright labour savings, but in the extent to which human expertise can be complemented with new technological capacities". Substitution is the lower-value strategy on its own terms.

What the skills evidence actually shows

Jobs and Skills Australia's national study looked at how skill demand is shifting rather than only at how many jobs might go. Its findings run against the assumption that AI capability is a technical training problem:

  • Generative AI "is increasing demand for both digital literacy and higher-order human skills like critical thinking, communication, and adaptability", at the same time, not in sequence.
  • It is "intensifying the pace of occupational skill evolution", which means the half-life of a given training investment is shortening.
  • Lifelong learning becomes a condition of employability rather than an enrichment activity, because job content is changing faster than it used to.
  • Skills development has to be inclusive, because cohorts are affected differently by occupation, industry and demographic factors, which means uniform programmes will leave the most exposed groups behind.

The pairing of digital literacy with critical thinking is the part organisations most often get wrong. Training that teaches people to prompt a tool without teaching them to evaluate its output produces confident users who cannot tell when they are being misled. That is a governance failure, not just a skills gap.

Redeployment is already the observed pattern

It is easy to treat upskilling as an aspiration. JSA's evidence from Australian early adopters suggests it is already the dominant response: "most observed impacts involve the evolution of roles, upskilling, and redeployment of workers, rather than widespread job loss". The study records reduced demand in some narrow occupations, but its overall finding is that "large-scale job displacement is yet to emerge, with current impacts limited to early-adopters".

Two readings of that are available. One is complacent: nothing much is happening. The other is that the organisations furthest along have generally found redeployment more workable than replacement, and that this is information about what works rather than merely about what is early.

Mobility is the harder problem

Exposure gets the attention; mobility determines the damage. JSA's warning is that "some roles face repeated exposure to automation with limited mobility options", and that workers' ability both to adapt within their occupation and to move between occupations "will be critical".

This is why aggregate reassurance is insufficient. A labour market where 79% of workers sit in low-automation-exposure occupations can still produce serious harm concentrated in the minority who are highly exposed and have few routes out. The policy and organisational response has to be targeted at that group, which requires knowing who they are rather than reporting an average.

JSA also notes that occupations experiencing faster skill evolution "tend to reward experience differently". Where the skill content of a role turns over quickly, accumulated experience is worth less relative to current capability, which changes the position of long-tenured staff in ways that seniority-based workforce planning does not anticipate.

Entry-level work is the pressure point

The most common worry about generative AI and careers is that it removes the bottom rung: if junior work is routine and routine work is automatable, the training ground disappears.

JSA's finding is more nuanced. Entry-level roles "may be more likely to transform than diminish", with no current evidence of widespread displacement in Australia, though the study notes this may partly reflect how early adoption is. What it does expect is that entry-level roles "are also likely to evolve, requiring more judgment and oversight of AI-generated outputs".

That is a demanding shift. It asks junior staff to supervise output in domains where they have not yet built the expertise that supervision requires. An organisation that removes routine junior work without redesigning how judgment is developed will find the gap several years later, at the point where it needs mid-level people it did not grow.

What this means for how capability is built

If adaptation is a determinant rather than a nicety, it belongs in workforce planning with the same seriousness as headcount. A few implications follow from the evidence rather than from preference:

  • Target it. Exposure is unevenly distributed, so uniform training under-serves the people who most need it. Identify the roles with high automation exposure and low mobility first.
  • Teach evaluation, not just use. The skill in demand alongside digital literacy is critical thinking. Training that stops at operating the tool leaves the risk in place.
  • Design for a shorter half-life. If the pace of skill change is intensifying, a one-off programme decays. Build a mechanism that updates rather than an event that happens.
  • Protect the development pipeline. Where routine junior tasks are automated, the judgment those tasks used to build has to be developed some other way, deliberately.
  • Watch the equity dimension. The ILO's finding that exposure concentrates in occupations disproportionately held by women in high-income countries means an untargeted programme can widen a gap rather than close it.

For a practical treatment of how this is structured inside an organisation, see our guidance on AI workforce capability uplift and on training employees as a governance control.

This article summarises published research on labour market exposure and skills demand. It is general information about workforce strategy and does not constitute legal, employment or professional advice. Decisions affecting individual employees carry obligations under employment, discrimination and privacy law in the relevant jurisdiction, and should be taken with qualified advice.

Related reading

Sources: ILO Working Paper 140, Generative AI and Jobs (20 May 2025) · Jobs and Skills Australia, Our Gen AI Transition (14 August 2025) · Jobs and Skills Australia, Our Gen AI Transition Analysis Papers (2 September 2025)