A dangerous myth is gaining traction with teenagers picking their university subjects, which I’m sure has also spread to many parents: there’s no point studying computing any more, because Generative AI can do it all now.
On results day, figures from UCAS showed that the numbers of students accepting places on Computing degrees at UK universities had fallen for the second year running; (by seven per cent for UK 18 year olds).
While we also know that people taking dedicated degrees in AI have been rising, these represent a very small minority of all computer science graduates.
My professional body, BCS The Chartered Institute for IT, is asking government, employers and my sector – education – to take this trend very seriously.
What a computing qualification really means
Why? Not just because of the risk of expecting whole areas of critical infrastructure to run safely on vibe-coded software, without expert technologists who understand what it takes to maintain it
The reason is that it’s built on a misunderstanding of what a computing qualification has ever been for.
A degree in computer science was never about learning a particular programming language, framework, or paradigm, any more than chemistry has ever been about mixing things in test tubes. The subject is and always has been about understanding the world around us well enough to make new things that work, and improve the things that don’t. In computing, that understanding has a name: computational thinking, the ability to take a messy real-world problem and work out how to make a machine solve it. Languages and frameworks come and go. That skill does not.
I suspect the myth comes from a reasonable-looking but false analogy: you can type a sentence into an image model and get a passable picture for a funny birthday card, so it feels as though you ought to be able to type a sentence or two into some other AI chat thing and get a working system of any complexity out of the other end. You cannot. Building, debugging, securing and maintaining real software is nothing like ordering a cartoon.
Evolving at pace
Generative AI is a genuinely powerful addition to the toolbox; it can help you think, save you time, point you at problems you would have missed, but it is a tool alongside the others, not a substitute for knowing what you are doing.
There is something particular about this technology that makes the panic understandable though. The past fifty years or so have not been short of disruptive technologies, and every one of them left computing a solid career. But the earlier ones were at least explainable. The computer: a machine that follows simple instructions mindlessly, but very fast. The internet: a way for computers to talk to one another. The web: a way to publish and link content. Each explanation fits neatly into a short sentence. We also saw these technologies evolve from their most basic form to what we know now. Generative AI, on the other hand, seemed to arrive out of nowhere in a startlingly mature form.
It keeps evolving at pace, and it is annoyingly and stubbornly resistant to explanation. I have spent more than a year trying, with colleagues who know far more than I do, to find a way to describe how it does what it does that is both true enough and clear enough to be useful to a general audience. I have failed. When even the practitioners struggle to explain the magic, it is no wonder the creative world is anxious about what it means for them, and no wonder computing is asking the same question of itself.
Lessons from a history of disruption
We have been here before though, even if in a smaller way. What we ask computers to do has changed beyond recognition over the past few decades. COBOL code to batch-process a payroll was miraculous in its day, but is orders of magnitude simpler than the infrastructure behind a single modern website. No sensible person would build all of that from scratch now.
As the work grew more complex and computers became more powerful, we kept finding ways to factor out and standardise the boring, repetitive parts, so we could concentrate on the creative, research-driven, entrepreneurial and business-critical ones. It would be unfair to generative AI to call it just the latest of these – it is more than that – but there is a family resemblance. It will not have the next world-changing idea for you. It will help you build that idea, if and only if you know what you are doing. It will help you critique it and debug it. And when the simple prompt “fix my code” stops working, it is the person who actually understands digital systems who becomes irreplaceable.
New paradigm, old skills
Which is the whole point. The most valuable thing a computing education gives you is not fluency in language X or framework Y. It is the ability to reason about how real-world concepts map onto the digital world, to think computationally, and AI raises the value of that skill rather than lowering it. The labour market is already saying so: demand for people who genuinely understand systems is climbing even as routine coding is automated away.
So the honest message to that sixteen-year-old is the opposite of the myth.
More and more people now need to understand enough of this discipline to use the new tools well, and to recognise when they are being led astray. And because AI and digitisation are reshaping every corner and crevice of society, we need more people, and a wider more varied range of them, going deeper into it, not fewer.
Writing code was always the easy part.
Computing is no more about computers than astronomy is about telescopes, so generally the right call.
And the same goes for language degrees – they’re about learning to communicate sensitively across cultural boundaries, not about being able to use the subjunctive, as much as I enjoy teaching grammar. Solidarity!
The problem with this argument is that generative AI doesn’t just know how to write code (“the easy part”). You’re describing the state of the world circa maybe 2024. The computer science tasks that you think humans will excel at working in conjunction with subservient AI—well, bad news, AI can now do that part too. Pay attention to what’s happening in pure math since late 2025.
The pure maths developments are truly amazing, but how much is that costing? ‘Sticker shock’ is already happening with LLMs in industry. These deep thinking maths models are economically a very long way from replacing humans in the kind of workplace decision making described in this article…
I disagree, you will always need humans to understand, if you truly want to rely on a system, what the code they have written/pushed/committed, does. If the humans who produce products and services with the code they write, don’t understand and safely verify the code they have written does what they need it to do, versus additional things or blind spots – hint: LLMs are probabilistic models, not deterministic. There will always be error bars in the output they produce. Use more critical thinking Chris.
Humans are incapable of understanding large human-written codebases, let alone verifying correctness. Because of that fact, Claude Mythos is literally too dangerous to release to the public, according to both its creator and the US government. Frontier models understand human-written code (and its inherent security flaws) to an extent that humans literally cannot, such that the very existence of these models pose a national security threat. Look into the actual details of the HuggingFace lab leak. How well did the humans do in that scenario, on both sides?
“LLMs are probabilistic models, not deterministic. There will always be error bars in the output they produce.” Yes, unlike humans right? *eyeroll*
The issue with this article is the BCS which is an irrelevant organisation that has not moved with the times. How many employers asked for BCS accreditation from their candidates? How many computing courses are BCS accredited? How many post -92 offer courses that were validated 5 years ago and are out of date? There is a difference between computer science and computing. The first teaches the theory; the second teaches how to use software.
Computer Science (and actually arguably even more so IT degrees) are likely to be more relevant in their purest form in the age of AI.
CS has become far too much about coding (in both students minds and the choices made to respond to that) in the lasts 30 years. AI is coming for the coding and, if anything it left, will leave the big picture stuff for the humans. This is where a CS degree should really be anyway in terms of its content. But not this’ll be better aligned with its value, if that makes sense.
The market for people with these skills may shrink however, as the age of “boilerplate coding job for every graduate” age ends. So there will be balance, somewhere between “don’t do CS, we’re cooked” (as you’ll read all over Reddit) and “just as many people can keep doing CS” and “CS cannot change and we can still keep treating a CS degree like a coding boot camp”.
The one I’m struggling with, as someone who doesn’t teach CS, but a related topic in a university, is how on earth we can re-skill to teach at a time where budget cuts leave no time to do CPD in work time. I don’t know how to oversee a proper agentic workflow and it’ll take me a lot of time to learn that to teach it!!!
AI is not a silver bullet.It is a popular misconception. AI relies on clean data which in turn relies on operational IT (and occasionally on syntheric data) plus data science and which are never actually discussed in any CS sysllaus I have seen , and I have seen a lot.Another myth, popular in schools, is that IT = coding. If an IT project is represented as a 12 hour clock, starting at 9 a.m., coding appears about noon. Lack of knowledge of what happens outside, say noon to 1 p.m. is unacceptable. IT is not a series of unrelated topics; it has a synergy where topics overlap or depend on some level of knowledge of other topics.
I find this debate particularly interesting because I am currently pursuing a Master’s in Artificial Intelligence, specialising further in machine learning and statistical learning while also studying AI governance.
I think there is truth on both sides of this discussion.
It would be a mistake to tell students that AI is simply another coding assistant. That description is already becoming outdated. AI systems are increasingly capable of debugging, reasoning across code, generating tests and completing larger software engineering tasks. The economics of software development will change, and some routine and entry-level coding work will almost certainly be affected.
But I draw a very different conclusion from that than “there is no point studying computing.”
If AI increasingly performs the execution, understanding computation, statistics, systems architecture, security, model behaviour and governance arguably becomes more important, not less.
There is also an important challenge to the traditional idea of human oversight. We often say that a human must always understand and verify what the AI has produced. But what happens when an AI can reason across a system too large or complex for any one human to fully comprehend?
At that point, “human in the loop” cannot simply mean a person checking the machine’s homework.
The question becomes: what must humans understand in order to remain meaningfully accountable for systems whose individual operations may exceed their own cognitive capacity?
That is one reason I chose to deepen my technical study of AI rather than move away from it. I do not necessarily need to compete with an AI over who can produce code faster. I need enough technical depth to understand what these systems are doing, interrogate their outputs, recognise their limitations, evaluate risk and participate competently in decisions about how they are deployed and governed.
So I agree with the central argument of this article, with one qualification: AI may indeed reduce demand for some forms of computing work. We should be honest with students about that.
But automation of coding is not the same thing as obsolescence of computer science.
If anything, the dangerous future would be one in which increasingly powerful computational systems are everywhere, while fewer humans understand the principles on which they operate.