The most common question I’m asked by parents and students has a surprising answer: yes, but not for the reasons you think, and not the version being taught today.
Every few weeks, a parent asks me the same question. Sometimes it comes at a family gathering, sometimes after a conference talk, sometimes in a LinkedIn message from a stranger: “My son is doing his A/Ls. Should he still do computer science? Won’t AI take all those jobs?”
It is the right question, asked for the right reasons, and it deserves a straight answer: not the reassurance the industry usually offers, and not the doom the headlines sell.
My answer is yes. Learn computer science. But understand what you are actually signing up for, because the profession you are entering will not resemble the one your seniors joined, and the version of computer science that most institutions still teach is preparing students for jobs that are quietly dissolving.
The job that is disappearing is not “Software Engineer”; it is “the specialist cog”.
For fifty years, software was assembled like a factory product. Requirements were written, handed to designers, handed to front-end developers, handed to back-end developers, handed to testers, handed to operations. Each handoff was a translation, and each translation lost meaning. We built entire org charts, and entire university curricula, around these silos. Front-end. Back-end. QA. DevOps.
Here is the uncomfortable truth: those silos existed because of human cognitive limits. No single person could hold the full stack in their head and stay productive. So, we divided the work in the name of specializing and increasing efficiency, but a different time is upon us now and we must act accordingly.
AI removes the reason for the division. When an engineer works alongside capable AI agents, one person can carry a product from intent to production, designing, building, testing, and deploying, holding the full context the whole way through. At Aventude, we call these people omni-stack engineers, and we have restructured our own delivery around small pods of them working with AI agents as peers, under a framework we developed called AISDLC.
The results are not theoretical. For a Stockholm-based construction logistics client, work conventionally estimated at two and a half years will now be delivered in roughly six months, not by working harder, but by removing the assembly line entirely. That compression is now the baseline against which global clients price software delivery. So, the honest career advice is this: the market will need fewer people who can only do one slice of the work, and it will pay a premium for people who can own the whole.
But is AI actually replacing engineers? The data says something more interesting.
If you only read headlines, you would believe AI writes flawless software and engineers are obsolete. The independent evidence tells a more useful story.
A randomised controlled trial by METR found that experienced developers using AI tools believed they were working about 20% faster, while the measured reality was quite different. The perception gap between how productive AI makes us feel and how productive it makes us is real and dangerous. GitClear’s analysis of hundreds of millions of lines of code found that AI-assisted codebases show record levels of duplicated code, with refactoring, the disciplined craft of improving existing code, collapsing to historic lows. Google’s DORA research found that AI adoption improved code quality at the micro level while reducing delivery stability, because AI lets teams ship larger changes faster, and larger changes break more things. And in developer surveys, the single most cited frustration with AI is code that is “almost right, but not quite,” which is more expensive to debug than code written from scratch.
Read that evidence carefully and you will see what I see: AI does not eliminate the need for engineering judgment. It concentrates it. When a machine can generate ten thousand plausible lines of code in a minute, the scarce skill is no longer typing the code. It is knowing whether the code is right: whether the architecture will hold, whether the requirement was even the correct one, whether the “almost right” answer will quietly corrupt a business process six months from now.
That judgment is precisely what a real computer science education builds. Data structures, systems thinking, complexity, trade-offs, correctness: these were never about syntax. They are the foundations of taste, and taste is what separates the engineer who directs AI from the operator who is directed by it.
Software won’t be built. It will form.
At Aventude, and through Skarpa, the AI-native future work platform, we operate on a thesis that sounds radical until you watch it in practice: software is stopping being a manufactured artefact and becoming something closer to a living system. Business intent goes in; understanding forms; working software grows around an organisation’s accumulated knowledge, continuously, accountably, with humans and AI agents sharing responsibility rather than one merely operating the other.
In that world, the valuable human is not the one who memorised a framework that will be obsolete in eighteen months. It is the one who can absorb a messy business problem, structure it, set the guardrails within which agents operate, and exercise final judgment over what ships. Deep specialists do not disappear; they are repositioned as the judgment layer: the architect who defines constraints, the security expert who audits, the domain expert whose knowledge the whole system learns from. That is a promotion, not a redundancy.
I will offer one honest concession, because credibility demands it: omni-stack does not mean uniform depth. There will always be moments that need a genuine specialist: a brutal database performance problem, a security review, a regulatory audit. The future team is T-shaped people whose horizontal reach is extended by AI, supported by a thin layer of deep experts.
What this means for Sri Lanka
Sri Lanka’s ICT and BPM sector earned an official US$1.65 billion in export revenue in 2025, with industry leaders arguing the real figure is considerably higher once offshore billing and freelance income are counted. It employs well over a hundred thousand people. It is one of the very few export industries that grew through every crisis this country has faced.
But much of that industry was built on a model the world is abandoning: selling engineering hours in silos. Headcount-based delivery. The body shop. If AI compresses a 2.5-year build into six months, a business model that bills by the person-month is not disrupted at the margins; it is inverted. The countries that win the next decade of software will be the ones whose engineers sell judgment, ownership, and outcomes, not hours.
That is a threat to the industry we have. It is an extraordinary opportunity for the industry we could build. An engineer in Colombo with strong fundamentals, communication and creative thinking skills, AI fluency, and end-to-end ownership skills can now deliver what previously required a team of eight in a Western capital. Geography has never mattered less. Judgment has never mattered more.
So, should you learn computer science?
Yes. But hear the fine print.
Do not learn it to become a “front-end developer” or a “manual tester.” Those titles are already melting. Learn it to become someone who understands systems deeply enough to direct machines that write code faster than any human ever will. Study the fundamentals hard: algorithms, architecture, data, security, because fundamentals are the only part of the curriculum that AI makes more valuable. Learn the business domains software serves, because translating human intent into working systems is the job that remains when typing is free. And treat AI tools not as an exam shortcut but as the colleague you will spend your entire career learning to lead.
The question was never really “should I learn computer science?” The question is whether you want to spend your career executing instructions or exercising judgment. AI has made the first path precarious and the second more valuable than at any point in the history of this industry.
Learn computer science. Then learn to be irreplaceable within it.