55% Fewer Graduate Jobs. AI Is Eating the Career We Told You to Study For.

A student asked me in October, at the end of a one-to-one tutorial session, something I've been asked in various forms by most of my S6 students this year. She'd been working hard, her mock results were improving, and she looked genuinely exhausted by the effort.

"Mrs. Lau," she said, "I just read that AI is replacing most entry-level analytical jobs. If that's true, what exactly am I working toward?"

I didn't give her a reassuring answer. I don't think reassuring answers are honest right now, and she's seventeen and smart enough to know when she's being managed rather than spoken to directly. I gave her the actual answer, which is longer and harder than "don't worry."

This is that answer.

The data is as bad as you've heard

In 2025, graduate vacancy postings in Hong Kong fell by approximately 55% compared to 2023 levels. Youth unemployment — the rate for 15-24 year olds — reached 12.3%, the highest in years. These figures aren't primarily a cyclical economic problem. They're structural. The employers cutting back on graduate intake are not planning to hire them back when conditions improve. They are restructuring toward workflows where AI handles tasks that previously required junior employees.

What's being displaced? The entry-level analytical work that was the first rung of professional careers in finance, consulting, law, accounting, and business: data collection and synthesis, first-draft research reports, document review, routine financial modelling, basic client communication management. These roles existed because human time was the cheapest way to do relatively structured cognitive tasks. AI changes that calculus.

The graduate who arrived at a financial services firm ten years ago learned the craft by doing those tasks — by processing documents, running models, making mistakes on low-stakes work, developing judgment through practice on real material. The pipeline from graduate intake to capable professional was built on that experience. Remove the entry-level layer and the pipeline is damaged in ways that aren't immediately visible but will compound.

This is not Hong Kong's problem alone. It's a global structural shift happening on an accelerated timeline. But Hong Kong's economy has specific vulnerabilities: concentration in finance and professional services, a labour market that has historically absorbed graduates well, a DSE-to-degree-to-job pipeline that a large proportion of families have oriented their children's entire education around. The disruption lands hard here.

What I tell students who ask "what's the point"

I tell them the question has two versions and they're conflating them.

The first version: "What's the point of DSE specifically?" This is a question about credentials and their market value. The honest answer is that a DSE certificate's labour market premium has always come primarily from tertiary admission, not from the DSE itself — and university degrees in fields with genuine analytical depth, clinical application, or human-facing specialisation still have value in a changed labour market. DSE is still the gate.

The second version: "What's the point of developing expertise and analytical capability when AI can replicate those capabilities?" This is the more serious question, and I think the honest answer is different from the one most careers counsellors are giving.

The capabilities AI is best at replicating are the ones that can be fully specified and trained on existing examples: classification, synthesis of structured information, template-based production, pattern recognition in well-defined domains. The capabilities AI replicates poorly are ones that require genuine novelty, relational trust, embodied presence, ethical judgment in contested situations, and the integration of sparse or contradictory information in high-stakes contexts.

The trajectory is this: AI gets cheaper and better at the former, continuously. The labour market premium for the latter increases, but so does the bar for demonstrating it.

Which skills are genuinely AI-resistant

Let me be specific, because generic "critical thinking" and "creativity" framing is unhelpful.

Domain expertise deep enough to evaluate AI outputs. The doctor who can assess whether an AI diagnostic suggestion is clinically appropriate is not replaceable by that AI. The engineer who understands the physics well enough to know when a model is producing physically nonsensical results is not replaceable. Expertise that can supervise and correct AI is more valuable than expertise that produces what AI now produces more cheaply. The implication is that going deeper in a domain — understanding not just the outputs but the principles — matters more than it did when shallow expertise was sufficient.

Judgment in genuinely novel or contested situations. AI is trained on past examples. Situations that are genuinely new — legal precedents without clear analogues, business decisions in unmapped territory, ethical questions where the relevant values are in tension — require human judgment. This is not a static safe harbour. AI will encroach here too, over time. But it's the direction to develop toward.

Complex relational and communication work. Negotiation, therapy, teaching, caregiving, leadership in ambiguous situations — these require something AI doesn't have: the ability to be genuinely present with another person, to respond to what they actually need rather than what the prompt specifies, to build trust over time through demonstrated reliability and care. The economy is already seeing premium wages for high-skill human-facing work. This will accelerate.

Synthesis across genuinely disparate domains. The most interesting work in the next decade will probably happen at intersections: between biology and computation, between law and emerging technology, between behavioural science and policy design. AI is excellent within a domain. Cross-domain synthesis that requires the human to hold multiple technical frames simultaneously and navigate between them is harder to automate.

Which degrees still have labour market value

I'm not going to pretend this question has a clean answer. Labour market conditions are changing faster than degree programmes can respond, and any specific claim I make today has a shelf life.

But directionally: degrees that provide deep technical or scientific foundations (medicine, engineering, computer science, applied mathematics) retain value because the underlying domains are expanding faster than AI is replacing practitioners in the upper tiers. Degrees in law and finance retain value conditional on specialising in areas where judgment, relationships, and novel situations dominate. Degrees in education, social work, and healthcare professions retain value because the work is fundamentally relational and embodied.

The degrees I'd be cautious about — and I say this to students directly, which makes some parents uncomfortable — are those where the core activity is producing standard-form documents, analysis, or reports for which AI already performs adequately. Business administration without a specific technical depth is a harder sell than it was. Generic media and communications is a harder sell. Not impossible — but the graduate entering these fields now needs a story about where they add value that AI doesn't.

Why DSE may need to be seen differently

The student who asked me in October is working toward a DSE result. She should work hard at it. The credential still matters for university access, and university access still shapes options in meaningful ways.

But I want her — and every S6 student I work with — to hold it differently than the previous generation held it.

DSE is not a finish line. It's not a promise. It's a demonstration, under conditions where external tools can't substitute for your own capabilities, of what you can actually do. In a labour market where AI can produce what many graduates used to produce, the premium goes to people who can genuinely do things — who have deep knowledge, genuine judgment, real human capabilities. DSE, whatever its limitations, tests those things in a way that ghost-written coursework and AI-assisted drafting don't.

The point of the work is not the certificate. The point of the work is becoming someone who can actually do something that matters.

That's still worth doing. The conditions under which it pays have changed, and changed significantly. But the work itself — the genuine development of real capability — has not become less valuable. It has, if anything, become more scarce.