Is Coding Still Worth Learning for Teenagers in the Age of AI?

Quick answer

Coding is still worth learning for teenagers in the age of AI - but the reason has moved up the stack. AI can now generate working code from a plain-English request, so the value of memorising syntax has fallen sharply. What has risen is the value of understanding software: decomposing a messy problem into solvable parts, reading and judging code you did not write, and debugging when the machine is confidently, fluently wrong. The labour-market data backs this up rather than undercutting it. The US Bureau of Labor Statistics still projects software-developer employment to grow about 15% through 2034, and the World Economic Forum lists software developers among the roles with the largest absolute growth to 2030. The honest conclusion is neither "stop learning to code" nor "code is everything". It is this: coding fundamentals plus judgement remain genuinely valuable, the routine production work is being automated, and the durable advantage belongs to students who can direct, read and verify software - not merely type it.

Why this matters now

The question is live because a credible figure made the provocative case, and the headline travelled further than the nuance. In 2024 NVIDIA chief executive Jensen Huang argued that the decades-old advice - that every child must learn to code - had inverted. His line, widely quoted, was that "the programming language is human": AI, he said, now lets people instruct computers in ordinary language, so domain expertise might matter more than syntax. Coming from the head of the company whose chips train the models, it landed hard, and plenty of parents heard a simple message - don't bother.

That is where the evenhandedness has to start, because the counter-evidence is just as serious. AI-generated code is not self-certifying. It needs human review, debugging and precision; it produces plausible solutions that compile and still do the wrong thing, and someone has to know enough to catch that. Coding also teaches something broader than coding - systematic problem-solving, the discipline of breaking a vague goal into exact, ordered steps. And the jobs have not evaporated. The US Bureau of Labor Statistics projects software-developer employment to grow roughly 15% through 2034, far faster than the average occupation, and the World Economic Forum's Future of Jobs Report 2025 lists software developers among the roles with the largest absolute growth to 2030, even as it forecasts wide churn elsewhere.

The Australian commercial picture sharpens the stakes further, and it cuts against the "coding is finished" reading. The Tech Council of Australia and the federal government have set a target of 1.2 million tech workers by 2030, up from roughly 950,000 in mid-2025 - a shortfall of around 650,000 people - with tech vacancy rates running about 60% above the national average. The Australian Computer Society has warned that the trajectory is "not on track". Most of that growth is what the Tech Council calls "indirect tech": technical roles inside banks, retailers, miners and government, not just at software firms. A country that short of technical capability is not a country where understanding software has stopped mattering. It is one where the kind of technical capability in demand is shifting.

What the coding debate is really about

The debate is not really "code or don't code". It is a disagreement about which layer of the work AI absorbs and which layer it hands back to humans - and on that, both camps are partly right. The first 1-2 sentences here are the whole frame: AI is automating the production of code while raising the premium on the judgement around it, so the argument is about where a teenager's effort now pays off, not whether software literacy is obsolete.

Huang's camp is right that raw syntax fluency is depreciating. The tools - natural-language code assistants, app builders such as Lovable and Replit - genuinely let someone describe an outcome and get a working draft, which means the hours once spent memorising a language's grammar buy less than they did. Jobs and Skills Australia's 2025 report Our Gen AI Transition reaches a compatible conclusion across the whole labour market: generative AI augments more roles than it replaces, and it lifts demand for digital literacy and human skills - problem-solving, communication, adaptability - rather than for narrow technical execution alone.

The critics are right that the judgement layer is now the bottleneck. Someone still has to specify the problem precisely, read the generated code critically, test it against real conditions, and find the bug the model introduced with total confidence. That is not a lesser skill than coding by hand; in many settings it is a harder one, because it requires understanding the system well enough to know when a plausible answer is a wrong answer. This is the same pattern McKinsey documents across enterprises in The State of AI 2025: adoption is nearly universal - 88% of organisations report using AI - yet only around 7% have fully scaled it to capture real value. Access to code-generating AI does not create capable software teams any more than access to a chatbot creates capable thinkers. The judgement to direct and verify is the scarce input.

Where coding genuinely still pays off

Used as a thinking discipline rather than a typing one, coding develops capabilities that survive the automation of syntax. The lead claim: the lasting return on learning to code is not employment as a typist of code but the mental architecture - decomposition, precision, debugging - that the practice builds and that AI cannot hand a student who never built it. Three worked examples show the value sitting upstream of production.

• The app builder. A teenager uses a natural-language builder such as Lovable to generate a working first draft of an app, then restructures the messy parts themselves. How AI assists: it produces scaffolding in minutes. What the student must verify: that the generated code actually does what was asked, by reading and testing it. The learning outcome: they learn to direct and judge software, not just accept it. The control: nothing ships until they understand how it works.
• The debugger. A student is handed AI-generated code that compiles but behaves wrongly, and has to find the fault. How AI assists: it wrote the draft and can suggest fixes. What the student must verify: which suggestion is actually correct, by forming a hypothesis and testing it. The learning outcome: systematic problem-solving - the transferable core of programming. The control: the bug is not "fixed" until the student can explain why.
• The decomposer. A student breaks a vague brief ("make a study planner") into exact, ordered, buildable steps before any code is written. How AI assists: it can implement each well-specified step. What the student must verify: that the steps are complete and correctly sequenced. The learning outcome: the single most transferable habit coding teaches. The control: the thinking - the specification - stays with the student.

In each case the value is upstream of production. The teenager who has written and judged real code can command a code-generating model; the one who has only ever prompted is at the mercy of whatever it returns.

Where the old case for coding no longer holds

Honesty cuts both ways, and several traditional arguments for "everyone must learn to code" have genuinely weakened. The lead sentence is the concession: some of the reasons parents were given a decade ago no longer hold, and pretending otherwise is how families end up over-investing in the wrong layer.

• "Coding is a guaranteed high-paying job." It is not a guarantee. Entry-level technical work is precisely the layer most exposed to automation, and the Australian Financial Review, drawing on Indeed Hiring Lab and Jobs and Skills Australia data, has reported graduate job postings falling around 15% in 2025 and noted that routine entry-level tasks are increasingly automatable. The bar for a first technical job is rising toward AI familiarity and data skills, not falling away.
• "Memorise the syntax." Syntax recall is the part AI does best. Drilling a teenager on language grammar for its own sake now buys less than teaching them to structure problems and judge solutions.
• "Pick the right language and you're set." Languages were always temporary; AI makes them more so. The durable layer is conceptual - how software is structured and reasoned about - not which framework is fashionable this year.

None of this argues against learning to code. It argues against learning to code for the old reasons. The reasons that survive are about thinking; the reasons that have faded were about production.

The Edison Method: where coding sits in Understand → Lead

At Edison AI Academy, coding is not the destination; it is one capability on the way to a larger one. The lead point: we sequence learning as Understand → Use → Evaluate → Build → Lead, and coding lives mostly in Build - valuable, but downstream of understanding and upstream of leadership, never the whole journey.

A student moves through the sequence rather than parachuting into the middle of it.

1. Understand - how software and AI actually work, including why generated code can be confidently wrong.
2. Use - direct tools, including code-generating AI, with clear intent and constraints.
3. Evaluate - read, test and judge what comes back, against real conditions and real sources.
4. Build - make working artefacts, where some coding fluency genuinely earns its place.
5. Lead - frame the problem, decide whether software is even the right answer, and explain it to people who will never read the code.

The point of the sequence is that Lead - judgement, framing, communication - is where the most valuable AI-era roles concentrate, and it sits above any language. This is why the better question is rarely "code or not", but "code as part of what". That broader question is taken up directly in should my teenager learn AI, coding, business or design, and the wider skill set is mapped in the AI skills students need before they leave school.

How to decide for your teenager

The decision is not binary, and treating it as binary is the main mistake. The opening guidance: give a teenager enough coding to understand and direct software, then weight their time toward the judgement, evaluation and communication that the market rewards above raw production - and let genuine interest, not fear, set the depth.

1. Start with literacy, not vocation. Every teenager benefits from understanding how software works the way they benefit from understanding how government or money works - as a citizen of a software-shaped world, not necessarily as a future developer.
2. Teach decomposition and debugging over syntax. Prioritise the transferable thinking - breaking problems down, finding errors - over memorising any one language.
3. Pair coding with evaluation. A teenager learning to code should learn, in the same breath, to test and judge AI-generated code. Production without judgement is the trap.
4. Let interest set the depth. A teenager who loves building should go deep; one who does not still needs the literacy, but their time is better weighted toward the broader capability. Forcing depth out of fear backfires.
5. Connect it to communication. The most valuable technical people can explain their work to non-technical decision-makers. Build that muscle alongside the code.

Common mistakes

• Hearing "the programming language is human" as "don't learn to code". Huang's point was about which layer matters, not that software literacy is worthless - and the jobs data plainly contradicts the literal reading.
• Over-investing in syntax. Drilling language grammar is optimising for the exact thing AI automates best.
• Treating coding as the whole answer. A student who can code but cannot frame a problem, judge output or communicate a solution has half the capability the market rewards.
• Treating coding as irrelevant. The opposite error: dismissing the decomposition, reading and debugging skills that have become more valuable, not less.
• Letting fear drive the decision. "Future-proofing" panic produces worse choices than genuine interest plus broad literacy.

How to know you have it right

The signal is not whether a teenager can produce code; it is whether they can think about software with judgement. The lead indicator: a well-prepared student can take a code-generating model's output and tell you where it is wrong, can break an unfamiliar problem into parts, and can explain the solution to someone who has never opened an editor. The poorly prepared one either cannot code at all and is mystified by the systems shaping their world, or can only ever accept what the machine returns because they never learned to read it.

The labour market is already pricing this distinction, and in Australia the stakes are concrete. PwC's 2025 Global AI Jobs Barometer found roles demanding AI skills carry a 56% wage premium - more than double the prior year - and continued to grow even as overall postings fell. Stanford's AI Index 2025 reports company AI adoption near 78%, which means the judgement to direct these tools is becoming a baseline expectation rather than a specialism - the broader capability set out in what AI education actually is. UNESCO's AI Competency Framework for Students (2024) makes the same case pedagogically, sequencing a human-centred mindset and ethics ahead of the technical layer and treating AI capability as a progression in judgement, not a syntax course. The students who learn coding as a thinking discipline - and pair it with the broader literacy and human skills around it - are being handed the advantage. The ones drilled only on syntax are training for the one part of the job that is already being automated.

So the recommendation is calm and specific. Do not pull your teenager out of coding because a chief executive made a clever line about human language, and do not pour their time into memorising syntax either. Teach them enough to understand and direct software, weight their effort toward decomposition, evaluation and communication, and let real interest decide how deep they go. Get that balance right and coding becomes what it should have been all along - not a vocational lottery ticket, but a way of learning to think clearly in a world that now runs on software it no longer makes you type by hand.