AI Is Replacing the Jobs We Trained Our Children For. Here Are the Skills That Aren't Replaceable.

By Mr. Ng · 15 November 2025 · 5 min read

I'm going to start with a specific scene, because I think it makes the abstract concrete.

A large financial services firm in Central recently restructured its graduate intake programme. They didn't eliminate it — they redesigned it. Roles that would previously have gone to eight fresh graduates now go to two. The six other positions? They were doing things that the firm's AI systems now do faster, with fewer errors, and at a fraction of the cost. The roles were not bad roles, or stupid roles. They were the standard entry-level analytical positions that Hong Kong universities have been training graduates to fill for twenty years.

This is not an isolated case. It is the pattern.

What AI Is Actually Displacing

Let me be specific, because vague statements about "AI replacing jobs" do not help parents make decisions.

Entry-level data analysis and report writing. The junior analyst who collects data from multiple sources, runs descriptive statistics, formats them into slides, and writes the narrative sections — this is now largely automatable. The tools exist, they are deployed, and they are faster and more consistent than a 22-year-old doing the same task for the first time.

Basic legal and financial document review. First-year work at law firms and accounting practices has historically involved ploughing through contracts and financial statements looking for specific items. AI performs this task with high accuracy and in a fraction of the time. The graduate programmes at these firms are contracting accordingly.

Standard coding tasks and quality assurance. Junior developers writing boilerplate code, testing simple functions, and doing first-pass debugging are being displaced. Not all coding — the design, architecture, and problem definition work remains human — but the implementation grunt work that junior developers used to learn on is automating away.

Data entry, coordination, and administrative synthesis. The range of roles that involve taking information from one place, transforming it, and putting it somewhere else — these are going quickly.

This is not speculation. Graduate vacancy numbers in Hong Kong confirm it: a 55% year-on-year fall in 2025, concentrated in exactly these entry-level analytical categories.

What the Research Says Isn't Going Anywhere

I've spent the last year reading the serious research on AI-resistant human capabilities — not the optimistic think-pieces, but the actual labour economics literature. Here is what it identifies, and why.

Genuine creative problem-solving — not the decorative kind. There is a category of creative work that AI performs well: generating variations, combining existing elements, producing content in established styles. But genuine problem-solving — defining what problem actually needs solving, in a novel situation, without a template — remains stubbornly human. AI is trained on what humans have done before. When the situation is genuinely new, it has less to offer.

This skill is built through genuine intellectual challenge, not drilling. The student who has only ever practised problems with known solutions is not developing this capacity, regardless of how high their grades are.

Complex interpersonal judgment. Understanding what a specific person actually needs in a specific moment, navigating trust, managing genuine conflict, persuading someone who does not want to be persuaded — these require theory of mind, emotional intelligence, and embodied social experience. AI can simulate the surface of these interactions; it cannot perform the underlying judgment reliably.

What builds this? Unscripted social experience. Debate. Sports. Theatre. Genuine leadership roles. The activities that Hong Kong education tends to treat as extracurricular decoration are, paradoxically, building skills that the academic curriculum is not.

Embodied and physical skills. Medicine, surgery, physiotherapy, skilled trades, engineering that requires physical inspection — these combine cognitive and physical capabilities in ways that are very difficult to automate. Robotic surgery exists; it is a surgical tool, not a replacement for the surgeon's judgment. The student considering medicine on purely credential grounds should know: the credential still opens a door, and what's behind that door is AI-assisted but still fundamentally human work.

Novel research and genuine synthesis. Not literature reviews — AI does those well — but the ability to identify what is worth studying, to ask a question that hasn't been asked, to synthesise across domains in a way that produces insight. This is graduate-level work at its most human, and it is the part of academic and scientific careers that AI has not touched.

What This Means for What You're Investing In

Here's the practical question: given all of this, where should parents actually be putting their educational investment?

Invest in depth over breadth. A child who understands mathematics deeply — who has really grappled with how and why it works — has more genuine analytical capability than a child who has been drilled in procedures across six different subjects. The depth of understanding, not the number of subjects covered, is what transfers.

Take interpersonal activities seriously. Stop treating debate, drama, and student leadership as "nice extras" for JUPAS applications. They are building the skills that the market is going to reward most highly in fifteen years. A child who can lead a room, think on her feet, and navigate genuine disagreement is building something AI cannot replicate.

Prioritise genuine curiosity over compliance. The student who asks "but why does that work?" — who won't accept a formula without understanding it — is developing a cognitive habit that is deeply resistant to automation. That habit is fragile. It requires adults who reward the question rather than just the right answer. Most tutoring does the opposite: it is optimised for correct answers, not genuine understanding.

Language and communication remain deeply human. Not basic writing — AI writes fluently. But the ability to understand an audience, to choose the register and approach for a specific person in a specific situation, to write something that lands emotionally as well as informationally — this is a rare human skill that the AI-saturated workplace will value more, not less.

The Honest Conclusion

The education system my generation inherited was largely a credential-sorting machine. Work hard, collect the right credentials, enter the appropriate career pipeline. That machine is breaking down, at least for certain categories of work.

What replaces it is not absence of effort or qualifications. It is a different orientation: building genuine capabilities rather than collecting performances. The distinction is real. A student who has genuinely learned mathematics can use it flexibly; a student who has been tutored to produce correct DSE answers has a credential, and the credential is worth less every year.

The shift required is less about which subjects children study and more about how they study them. That's the conversation Hong Kong's education system needs to have with itself.