The great graduate divide
Graduate positions have long been mooted as the most desirable jobs on the market for young people. Seen as the gateway to long-term career success, graduate jobs have always been competitive, but the widescale adoption of AI has seen many of these roles become obsolete.
According to the High Fliers Research Graduate Market Report for 2026, graduate recruitment within the UK’s leading employers has slumped by 24.5% since 2022 – a steeper fall than either the 2008 financial crisis or the COVID pandemic. Since 2024, graduate vacancies have dropped by nearly 20%, despite a 1% increase in undergraduate students.
The data calls this an employment issue, but I think that’s only half the story. There’s another story, which focuses on workplace psychology, and it’s one that most leaders aren’t equipped to tell.
In the past five years, AI has gone from a futuristic, sci-fi-esque concept to a widely adopted business operations tool. Whilst for many businesses, that has improved efficiency and reduced overheads, it has created a talent development gap.
AI is extraordinarily good at absorbing the tasks that junior employees used to undertake. First drafts, research, admin – the list goes on. The output of these tasks was always important, but it was also the mechanism through which young people built competence, confidence and a sense of belonging at work.
By removing that mechanism, businesses haven’t just removed career pathways, they have removed the developmental scaffolding of a generation that was learning to become leaders.
This idea builds on a theme I’ve explored throughout my work: growth has never come from comfort; it comes from productive discomfort: the not-yet-knowing, tested against real stakes and survived. Junior roles were never really about the output; they were a sanctioned space for that discomfort, low enough stakes that getting it wrong didn’t end a career, but real enough that getting it right built genuine belief in one’s own capability. When AI absorbs the task, it doesn’t just absorb the output, it can quietly absorb the discomfort too, and with it the very mechanism through which resilience and self-belief are built.
The grunt work was never just grunt work
Leaders often forget that career development isn’t just achieved through the completion of tasks or gathered experience. It is also achieved within the realms of psychological safety – an environment that allows junior professionals to learn without fear of consequence, one small, correctable mistake at a time. Completing tasks develops one set of knowledge, but the understanding of process, the development of relationships and social understanding, is just as important for development.
Artificial intelligence often completes those same tasks faster, and often better, than its human counterparts. techUK’s analysis suggests that 50-60% of junior-level tasks have been absorbed by automation, and particularly effective at increasing the output of the least experienced staff. Whilst this is positive news for a company’s bottom line, it does leave organisations vulnerable to a psychological skills gap in the years to come.
This element gets lost in the productivity narrative. If people aren’t given the opportunity to make low-stakes mistakes, they fail to develop what Albert Bandura called self-efficacy; the belief that they are capable of doing hard things. That belief is anchored in success, but also in the experience of managing failures: missed deadlines, a fractured internal relationship, incorrect research. If AI removes the need for human input, then we will develop technically fluent leaders who have never actually failed, and therefore cannot lead from a position of psychological safety.
Psychological safety was built into the old apprenticeship model
Amy Edmondson’s research on psychological safety tells us that people take the interpersonal risks that are essential to learning (asking a ‘stupid’ question, admitting confusion or asking for help) only when they trust that doing so won’t damage how others see them. The traditional entry-level job created this safety by default, as junior work was intentionally low stakes as employers understood that people are imperfect, and as a result make imperfect decisions.
Apprenticeships created psychological safety, by positioning the roles as ‘on the job learning’. Employees were given immediate freedom to be honest about their skills gap, with the core understanding that their sole purpose was to develop. It’s no wonder that the latest figures suggest 85% of apprentices will go on to secure a full-time role after completing their apprenticeship, as opposed to 68% of university graduates.
What deliberate development now requires of leaders
None of this is an argument against AI – it is clearly a beneficial and necessary tool for modern day businesses. NACE’s spring 2026 outlook shows demand for AI skills in entry-level roles has nearly tripled since autumn 2025, largely because it’s shifting junior work toward analytical, higher-value tasks.
The opportunity is real, but can only materialize if leaders actively redesign the developmental path AI has disrupted, rather than assuming it will reassemble itself. Four behaviours matter most.
First, leaders must replace lost repetition with structured stretch, and give juniors work which breeds accountability, so competence is still built by doing, rather than reviewing. Second, they must protect psychological safety deliberately and with intention: name the learning curve out loud, normalise questions in team settings and allow mistakes to be viewed as opportunities to learn, not be scolded. Third, they must rebuild the touchpoints that AI has removed – pairing juniors with senior leaders through something like a sponsorship scheme makes junior talent 53% more likely to advance to the next rung of the leadership ladder than their unsponsored counterparts. Fourth, leaders must resist using AI’s efficiency gains to justify not backfilling junior roles at all, or they risk developing a workforce that is lacking in psychological safety, in turn making them less productive and at a higher risk of leaving their position.
The ask
The resolution to this issue cannot be owned by one party – it is the responsibility of employers, educational providers and policy makers to ensure graduate talent pathways can co-exists alongside AI automations. Employers need to redesign entry-level roles around explicit ownership and sponsorship. Education providers need to build psychological resilience before young people arrive in the workplace. Policymakers need to keep apprenticeships central to employment infrastructure, not treated as ‘lesser than’ by employers.
The organisations that thrive over the next decade won’t be the ones with the most advanced AI, they’ll be the ones that got comfortable building discomfort back into how people learn to lead.
