Most of the conversation about AI and employment focuses on the middle: the office worker whose tasks are being automated, the knowledge worker watching a language model do in seconds what used to take an afternoon. That conversation is real. But there’s a different disruption happening at both ends of the career arc, and it’s getting far less attention.
At one end: a 2025 graduate applying for roles they’re qualified for, in fields they studied for, and finding that those roles no longer exist at scale. At the other end, a 57-year-old professional with three decades of contribution who finds their role restructured away at the exact moment when retraining is most costly, and re-entry is hardest.
These are not the same situation. But they share something structurally important: both groups are bearing costs they did not choose and cannot easily absorb, for decisions made in rooms they were not in, to benefit shareholders they do not represent.
That’s the conventional framing. But notice what it leaves out: whether this is merely the latest chapter in a long history of technological disruption, or something categorically different — and therefore requiring a different kind of response. I want to take that question seriously.
## The ladder has a missing rung
The data on entry-level hiring is striking. Entry-level job postings have declined by roughly 35% since January 2023, according to data from Revelio Labs reported by CNBC. A SignalFire study found a 50% decline in new role starts by people with less than one year of post-graduate experience between 2019 and 2024. Only 30% of 2025 graduates secured an entry-level role in their field, down from approximately 40% the year before.
But the number that keeps coming back to me is this one: a 73% average decrease in entry-level hiring rates, compared to a 7.4% decrease across all job levels in 2024–25. That’s not cyclical softness. That’s a structural shift concentrated almost entirely at the bottom of the career ladder.
Here’s what I think is actually going on. AI has not just automated entry-level tasks — it has automated the *apprenticeship*. The work that used to sit at the start of a career —data entry, report drafting, meeting notes, first-pass analysis — was never really about the output. It was about the formation. You learned how organisations made decisions by being asked to summarise them. You developed professional judgement by practising low-stakes versions of it. The tasks were the training.
That training is now being replaced by AI agents. And the new roles AI is creating — AI orchestrators, model evaluators, judgment-in-the-loop reviewers — assume the professional maturity that entry-level work used to build. They require seniority. This creates a paradox that no one seems to be solving: you need experience to get the new jobs, but the jobs that build experience no longer exist at the scale required.
This has a second-order consequence that I don’t think enough people are thinking about. If the pipeline into the profession is blocked, the supply of future senior talent starts to thin. Organisations may be, without realising it, weakening the very system that produces their next generation of decision-makers. It is too early to say this will happen.
It is not too early to ask whether anyone is planning for it.
## The contract that wasn’t written down
The situation for older workers is different in almost every respect, but it lands in the same place. Workers in their 50s and 60s entered careers under an implicit compact. It was never a formal contract — it was an understanding: contribute, develop expertise, advance, and the organisation will carry you through to the exit with your dignity intact. AI-driven restructuring is breaking that compact, quietly, and at scale.
The AARP’s 2025 research found that 61% of workers aged 50-plus are concerned about AI replacing workers, and 59% worried about displacement. Those numbers track with the structural reality: many of the roles most exposed to AI automation — document processing, reporting, mid-level analysis — are exactly the roles that experienced workers have built years of seniority in.
The more interesting question is not who is worried, but who is most vulnerable. The research suggests the risk is uneven. Workers in high-experience-premium roles — where tacit knowledge, relationship capital, and contextual judgement matter — may find AI complements their work rather than replaces it. The threat is most acute for experienced workers in AI-exposed, experience-light roles: the roles where the tasks, not the expertise around them, define the position.
Workers aged 55-plus are simultaneously the fastest-growing segment of the labour force and among the most poorly served by existing transition infrastructure. Access to employer-provided retraining is lower for this cohort. The cultural bias against older workers learning new technologies — which research consistently identifies as a barrier, even when the workers themselves are willing to adapt — compounds the structural problem.
Redundancy is lawful. No one is disputing that. But legality and ethics are different registers. The question of what organisations owe to workers who have given them decades of service is not answered by the Fair Work Act.
## Three tensions worth sitting with
I want to be careful here not to collapse these into a single narrative, because they are three separate problems running in parallel. The first is *intergenerational*. A generation of graduates is entering the workforce at precisely the moment when the entry mechanism has been disrupted. This is not about a few industries automating some roles — it is about the professional formation pathway being closed off at scale, across industries, for a cohort that had no say in it.
Stanford researchers found a 16% decline in early-career employment across AI-exposed occupations since late 2022. If the current graduating years — 2024, 2025, 2026 — are the most disadvantaged cohort of graduates in a generation, then the question of who bears responsibility for repairing the pathway is urgent and currently unanswered.
The second is *contractual*, in the broader sense. There is a reasonable argument that organisations which have extracted value from experienced workers for decades have obligations toward them that exceed the statutory minimum — not because the law requires it, but because that is what a person or organisation of good character would do. The fact that a company *can* restructure a 58-year-old out of their role without a meaningful transition plan does not mean it *should*.
The third is *distributional*. The productivity gains from AI automation are accruing, primarily, to shareholders and the organisations making the automation decisions. The transition costs — unemployment, retraining, lost income, the psychological weight of being discarded mid-career — are falling on workers who had no representation in those decisions. This is not a new dynamic in technological change. But scale and speed matter.
The rate of displacement in entry-level roles, in particular, is fast enough that the social infrastructure for managing transition has not kept up. These three tensions are genuine. I don’t think they resolve neatly, and I’m wary of anyone who tells you they do.
## The governance gap
Australia’s regulatory response to AI and workforce displacement is, at this point, principled and largely theoretical. The APS AI Plan, published in November 2025, outlines reskilling pathways and AI literacy programmes. An AI Safety Institute has been established. A gap analysis has been commissioned to test whether existing workplace laws are adequate for AI-related decisions.
These are real steps.
But Australia’s framework remains voluntary and principles-based. There are no mandatory transition obligations for employers whose AI decisions displace workers. There is no requirement for transparency — most workers do not know that AI was used in the decisions that affected their employment. There is no conformity assessment requirement for AI used in hiring or restructuring, despite age being a protected characteristic under the Age Discrimination Act 2004, and AI systems potentially producing discriminatory outcomes for older workers without intent or awareness.
The Australian Government’s own Voluntary AI Safety Standard — Guardrail 4 specifically — requires meaningful human oversight and control of AI in decisions that affect individuals. Whether that standard is being applied to workforce AI decisions in most Australian organisations is, at best, uncertain.
This contrasts with the direction the European Union has taken. The EU AI Act classifies AI used in employment as high-risk, requiring transparency, human oversight, and conformity assessments before deployment. The Finance Sector Union has argued that without enforceable rules, voluntary principles will not change corporate behaviour. The Business Council of Australia has pushed back, pointing to the EU as a cautionary example.
Both positions are coherent. Neither has resolved the underlying problem for the workers currently being affected.
The data here is real, but data rarely settles arguments like this — it just reframes them.
The reframing on offer is this: Australia is currently allowing significant workforce decisions affecting protected groups to be made by automated systems, with no transparency requirements and no audit obligations. Whether that is acceptable is a question the gap analysis will eventually have to answer.
## What we don’t yet know
The question I’d put to any leader making AI-driven workforce decisions right now is a straightforward one: do you know which of your current practices are creating these costs, and have you decided — consciously, not by default — that they are acceptable? That is not a regulatory question. It is a governance one. The fact that most organisations cannot answer it is itself an answer.
I want to close not with conclusions but with the questions I think deserve more attention. The first is the pipeline question. If entry-level roles continue to decline at the current rate, where does the next generation of senior talent come from?
No one in Australian business or policy seems to be addressing this at scale. It may be the most consequential long-term consequence of the current disruption, and it is almost entirely absent from the public conversation.
The second is the question of organisational obligation. When a company automates away roles that workers spent careers building competence in, what — if anything — does it owe those workers beyond the legal minimum? The current Australian framework provides no answer. Ethicists, employment lawyers, and business leaders are not having this conversation in the same room.
The third is the pace question. The rate of change in AI capability is outrunning both the policy frameworks designed to govern it and the social infrastructure designed to absorb the disruption. We have been here before — in manufacturing, in telecommunications, in financial services. The transitions were painful and slow.
The question worth asking is whether the current pace allows for a managed adjustment, or whether the speed itself is the problem that most needs addressing.
I don’t have clean answers to any of these. What I keep coming back to is this: the two ends of the career are carrying a disproportionate share of the costs of a transformation they did not choose, and the people making the decisions that produce those costs are not yet being asked to account for them.
That seems like a problem worth naming, even before we know how to solve it.
—
*Sources: Revelio Labs / CNBC (2025); SignalFire workforce study (2024); AARP Workforce
Trends report (2025); Stanford Digital Economy Lab, Brynjolfsson et al. (November 2025);
Goldman Sachs AI employment analysis (2025); APS AI Plan (November 2025);
Australian Government Voluntary AI Safety Standard — Guardrail 4; EU AI Act
(Annex III, employment AI); Age Discrimination Act 2004 (Cth); Fair Work Act 2009 (Cth).*
—
Leave a Reply