Should Kids Still Learn to Code in the Age of AI? Absolutely.

Introduction

In 2023, Andrej Karpathy remarked: “The hottest new programming language is English.” A year later, at the 2024 World Government Summit in Dubai, NVIDIA CEO Jensen Huang said:

“It is our job to create computing technology such that nobody has to program, and that the programming language is human. Everybody in the world is now a programmer.”

This statement sharply contrasted with what has been said for years from similar stages, that everybody should learn computer science and coding. Many governments have introduced programming into early education, and STEM has become a global priority.

I don't really know what is meant by "our job" in this quote, therefore who is the implied "we" there, so I won't focus on that part.

These recent claims have sparked an important conversation. Many people ask me what will happen to programming, mainly because they know I’m a software engineer and have experience in education. Some connect big-tech layoffs to AI, while aspiring developers feel uncertain about their future. YouTube is filled with videos declaring that “programming is dying.” I’d like to share my perspective because I’m in this field, and I think about these transformations every day.

A Short History of Programming Evolution

Programming has evolved through roughly five generations:

1. Machine code
2. Assembly language
3. High-level languages (C, Fortran, etc.)
4. Domain-specific high-level languages
5. Very high-level languages (Python, SQL, etc.)

Now, natural language is emerging as the sixth generation of programming, where you can generate working applications without explicitly writing code.

Each generation made it easier to automatically produce the previous one. Assembly made machine code simpler to write; high-level languages made assembly easier; and so on. In that sense, I agree with Karpathy: English is indeed becoming the new programming language. For me, this shift is neither unexpected nor frightening; on the contrary, it is an exciting development that I wholeheartedly welcome.

But there’s a crucial difference. Previous generations were deterministic. When you write a C++ or Python program, you don’t check the exact machine code that’s produced—you trust the compiler. Natural language, however, is probabilistic by nature. We want it to be flexible and fluid. Yet to achieve useful outcomes, we end up crafting structured prompts, which sit somewhere between natural language and traditional programming. Isn’t that why we call it prompt engineering?

What AI Can (and Can’t) Do Today

As of late 2025, I don’t know of any AI system capable of generating production-ready, customer-facing complex applications from pure natural-language prompts.

Most demos are to-do lists, small company websites, or prototypes. That doesn’t mean AI can’t generate sophisticated components, but no system today can produce an entire robust product without human guidance.

In my experience, it’s an iterative process—prompting, checking, refining. And to evaluate AI-generated code, you still need to understand programming.

I can’t imagine someone with zero computer-science knowledge producing a working application on their own. Many blog posts describe users happily building prototypes, but I’ve yet to see a large-scale production system built by someone who doesn’t understand code. Even if a 90% ready app is built, in production environments, 90% reliability isn’t enough. Even small inconsistencies can be catastrophic.

As a thought experiment, let’s imagine a super AI capable of doing absolutely everything. Could I simply write:

“Analyze company ABC’s industry and design and create software products that outperform ABC’s offerings in every dimension.”

—and have it executed perfectly? I don’t believe any company is building such an AI. In fact, it’s in the interest of existing AI producers to protect their market positions rather than create self-defeating products.

Now, let’s scale this down. Imagine a scenario where anyone could use AI to create a fully functioning platform-level app — something like Uber or Airbnb. If AI could build and operate the entire platform, it could indeed disrupt entire industries. But if AI only generates components of the system, we still need human tech professionals to integrate, maintain, and optimize the rest.

The first scenario is essentially a “Skynet” scenario, leaving little room for human activity because AI is all-encompassing. The second scenario, however, is much closer to reality today: AI assists humans, enhances productivity, and changes the nature of work rather than replacing it entirely.

The Future: Smarter AI, but Not Omniscient

Will we ever reach the point where AI systems can fully understand and build what we mean in plain language—with 100% accuracy? Possibly, but only up to a point.

Do we even want that? Sometimes, when I ask an LLM to generate something, it produces extra artifacts I didn’t ask for. Human intent is nuanced—and total implicit understanding can be risky.

Should Children Still Learn to Code?

My answer is absolutely yes.

Here’s why:

• Computer science is deep and complex. Processors still run on binary code. Even if we prompt in English, there are countless intermediate layers before instructions reach the hardware. We need structured determinism to ensure machines behave reliably.
• To understand, debug, and optimize AI-generated code, we still need knowledge of algorithms, data structures, operating systems, networks, and software design patterns. If everyone stops learning computer science, who will build and maintain the systems that enable this AI revolution in the first place?

When we say “learn to code,” we don’t just mean typing commands in an IDE. We mean building systems, solving problems, and architecting logic. Every coder is an architect of sorts.

If you enjoy programming and computer science, pursue it confidently—it remains a valuable, future-proof profession.

Domain expertise in other scientific fields is certainly valuable. Computer science is, after all, an advanced tool for transforming nearly every domain of human activity. However, having deep knowledge in another field doesn’t mean that computer science is a solved discipline or that there’s little left to learn. On the contrary, the rise of AI is opening entirely new frontiers in computer science, precisely because it accelerates how we explore and expand the field itself. For example, improving AI safety is one of the most critical fields today. Focusing on it is an existential priority for humanity.

Will IT Jobs Decrease?

In some areas, yes.

Many large SaaS companies already have mature, stable codebases. Their products dominate the market, and major innovation slows down. In these environments, AI will accelerate maintenance and incremental development, reducing the need for large teams.

But even without AI, such layoffs would eventually occur due to system maturity. AI is merely amplifying that trend, not causing it outright.

The New Demand: AI-Augmented Programming

AI will change programming jobs—but it won’t eliminate them.

In fact, a new kind of demand is emerging: programmers who can integrate and maintain AI systems, especially as agentic AI expands into every sector. Those with computer-science training are better positioned to design effective prompts, validate AI output, and build hybrid systems that blend logic with language.

Developers have always sought faster, more efficient ways to build software. That’s why most of us embrace AI—it’s “vibe coding,” not job destruction.

Final Thoughts

The essence of programming is shifting from writing code to expressing intent. But that doesn’t make programming obsolete.

There is no doubt that AI is democratizing programming, making it more accessible than ever—and that is an incredibly positive development. But this doesn’t mean foundational knowledge is obsolete. Just as we wouldn’t tell children they don’t need to learn music theory or how to read notes simply because apps can convert humming or whistling into full compositions, we shouldn’t suggest that coding fundamentals are no longer needed.

Observing that things just work is useful, but understanding why they work remains a core part of human learning and creativity. As Albert Einstein once said:

Coding is not dying; it’s evolving. The best way forward is to blend human creativity, structured thinking, and AI assistance.

If anything, this era reminds us that understanding how systems work, not just how to prompt them, will remain one of the most valuable skills in the world.