No Degree Is AI-Proof: Why Delayed Specialisation May Help Students

A recent survey of New Zealand business leaders found 87% of organisations have seen job roles change or disappear because of AI, with a third reporting slower entry-level hiring. Against that backdrop, two researchers argue in The Conversation that no university degree is immune, but that graduates who specialise too early may be most exposed. Their prescription: broader foundational years, deliberate cross-disciplinary learning, and AI treated as a core literacy rather than an elective topic.

What happened

Data point	Detail
Organisations reporting AI-driven job changes	87% of surveyed New Zealand business leaders
Organisations slowing entry-level hiring	One third of the same survey group
AI tasks already displacing graduate work	Drafting reports, summarising research, writing code, analysing data, generating content
Proposed graduate model	“T-shaped”: deep discipline expertise plus cross-disciplinary breadth

The pressure on university degrees is no longer theoretical. According to the survey cited in the piece, a large majority of New Zealand employers have already restructured roles because of generative AI, and junior positions are the first to feel it. Graduates entering the workforce in the next two to four years will compete for fewer entry-level openings at the exact moment AI can credibly produce many of the outputs those roles once required.

The authors are clear that the problem is not AI itself. Generative AI still cannot reliably judge whether its own output is accurate, responsible, or fit for purpose. That gap between producing an answer and evaluating it is where graduate value now lives.

Why does this matter for employers and businesses?

If you are hiring, the workforce pipeline is changing faster than most curricula. A graduate who spent three years going deep on a single discipline without any structured exposure to adjacent fields may struggle to work on the messy, cross-functional problems that make up most real business decisions.

The researchers point to consulting firms as a concrete example. Those firms already recruit from commerce, computer science, psychology, and journalism. What unifies successful hires is not the degree title but the ability to analyse unfamiliar problems, work across disciplines, and exercise judgement under uncertainty. These are also the capabilities most at risk of being hollowed out in roles that consist mainly of routine information processing.

The argument also applies to the AI tools your business uses today. A polished-looking AI output can conceal weak reasoning, missing evidence, or fabricated facts. Someone who can question, verify, and take responsibility for that output is more valuable than someone who can simply prompt the model. This is not an abstract point about education policy. It is a hiring and workflow problem you may already face.

What the T-shaped model actually looks like

The researchers propose a “T-shaped” graduate as the practical target. The vertical stroke is genuine discipline depth: an accountant who can defend an analysis, a lawyer who understands precedent, an engineer who can judge whether a design is safe. The horizontal stroke is the ability to communicate across fields, use AI critically, and make decisions under uncertainty.

In practice, they suggest universities could:

• Begin degrees with a broader foundation that pairs disciplinary content with AI literacy, ethical reasoning, and collaborative problem-solving.
• Move specialisation later, after students understand how fields connect.
• Create shared courses across faculties, for example business, design, and computer science students working together on responsible AI adoption, or law and health students examining privacy in automated decision-making.
• Redesign assessment to reward explaining reasoning and defending decisions, not just producing correct answers.

The last point is the sharpest one. As information becomes easier to generate, assessment that simply asks for an answer is no longer a reliable signal of competence. Universities that shift the bar toward reasoning and accountability will produce graduates who are genuinely harder to replace.

Our take

The 87% figure is striking, but it is a self-reported survey of business leaders, not an independent audit of headcount data. Take it as a directional signal, not a precise measure. That said, the underlying logic holds. We see it in the projects we ship: the bottleneck is rarely producing content or code, it is the person who can review it, spot the error, and decide what to do next.

The T-shaped framing is useful, but universities move slowly. If you are a student or a parent reading this, do not wait for your institution to redesign its programme. Actively seek cross-disciplinary electives, work on real projects outside your faculty, and treat AI tools as something to understand critically, not just use fluently. For businesses thinking about integrating AI into their workflows, the same logic applies: the value is in the human judgement layered on top, not the output itself.

The degrees and the workers who survive this shift will be the ones who can do what AI still cannot: decide whether the answer is any good.

What to do about it

1. Audit which entry-level tasks in your business are already being handled by AI, and identify where human judgement is still the bottleneck.
2. When hiring, weight interview questions toward reasoning and verification, not just domain knowledge.
3. If you manage a team, run a short internal exercise where staff review an AI-generated report for errors and hidden assumptions. It will reveal your real capability gaps quickly.
4. If you are a student, treat cross-disciplinary courses as a strategic investment, not a distraction from your major.
5. Talk to the Lumien team if you want a practical view on which AI tools are ready to use in a business context and which still need heavy human oversight.

Source: Bing News · Claude AI