It's Still Worth Learning to Code

"Why learn to code when AI can just write it for you?"
"Coding bootcamps are dead."
"Junior developers are obsolete."

You've heard this. Maybe you've started to believe it. Maybe you're halfway through a Python tutorial and wondering if you're wasting your time.

You're not. Anders Hejlsberg — the creator of C# and TypeScript, a Technical Fellow at Microsoft — thinks the same. "AI is an accelerant," he said in a recent interview. "It is not a replacement."

Here's why he's right.

The Argument Against Learning to Code (And Why It's Shallow)

The argument goes like this: AI tools like GitHub Copilot, Claude, and ChatGPT can generate working code from a sentence. Why spend months learning syntax, debugging logic errors, and memorizing APIs when a prompt can do it in seconds?

It sounds compelling until you try to use those tools on a real problem.

Hejlsberg draws a sharp distinction here: AI excels at "grunt work" — writing unit tests, generating boilerplate, producing a to-do list app it has seen a million times. But it cannot innovate. It cannot create fundamental designs from scratch. The moment you hit something non-trivial — a race condition, an unexpected API behavior, a performance issue under load — the AI gives you plausible-looking code that doesn't quite work. And if you don't know enough to evaluate it, you're stuck. Not just stuck — confidently stuck, which is worse.

Knowing how to code doesn't mean you'll outtype the AI. It means you can tell when the AI is wrong.

What Coding Actually Teaches You

The surface skill is syntax. The real skill is something harder to name.

Hejlsberg has a blunt diagnosis of where beginners go wrong: they focus on the "syntax veneer" instead of understanding semantics, data structures, and how things like pointers or memory actually work. The syntax is just notation. The underlying concepts — how a computer executes, why things fail, what abstraction is doing — that's what sticks.

Learning to code forces you to build a precise mental model of how machines execute instructions — how memory works, how functions compose, why order of operations matters, what "asynchronous" actually means. That model is what lets you reason about a system you've never seen before.

It's the difference between reading a map and having internalized a sense of direction. Both can get you to your destination. But one of them works when the GPS breaks.

When you've written a sorting algorithm by hand — not because you'll ever use it in production, but because the exercise forced you to think step by step about comparisons, swaps, and loop conditions — you carry something that no amount of prompt engineering gives you.

You carry the habit of exact thinking.

AI Makes the Skill More Valuable, Not Less

Here's the counterintuitive part: the more AI can do, the more valuable a real understanding becomes.

Hejlsberg makes a point that's easy to miss: someone still needs to design the CPUs, the operating systems, and the programming languages that AI uses to express itself. The AI doesn't design its own tools. It uses tools humans built — and humans who understand those layers are the ones who keep pushing the frontier forward. The floor of what AI handles keeps rising, but so does the ceiling of what's possible for someone who understands the foundation.

Ten years ago, a junior developer could add value just by writing straightforward CRUD code. AI can do that now. So the floor moved up. The work that humans need to do is now higher up the abstraction stack — reviewing architecture decisions, catching edge cases, understanding why a system is behaving unexpectedly.

All of that requires the mental model that only comes from having built things yourself.

"AI can write code. It can't build the judgment to know when the code is wrong."

The developers who thrive with AI tools aren't the ones who know the best prompts. They're the ones who can read the output critically, spot the flaw in the logic, and direct the AI toward something better. That skill is built through years of making your own mistakes — not through prompt-and-accept workflows.

"But I'll Never Be a Professional Developer"

Good. That's not the only reason to learn.

Coding is one of the few skills that lets you build tools for your own problems. Not someone else's product roadmap — your problems, on your timeline.

The data analyst who can write Python scripts to automate their own reports. The founder who can prototype their idea before they have a budget to hire engineers. The product manager who can read a pull request and actually understand what's changing. The researcher who can scrape and process data without waiting on a team.

These aren't people who need to be senior engineers. They need enough to be dangerous — and that threshold is absolutely achievable.

How Much Do You Actually Need to Learn?

Not as much as you think.

You don't need to memorize algorithms for a whiteboard interview. You don't need to understand every layer of the stack. You need enough to:

Understand what a program is doing when you read it
Debug something when it breaks
Know which questions to ask the AI (and when to distrust the answer)
Build small things end-to-end

That's maybe 6 months of consistent effort. Not mastery — enough. And with AI as a learning assistant, you can get feedback on your code instantly, ask questions at 2 AM, and move through concepts faster than any previous generation of learners.

Hejlsberg's advice for getting hired is telling: write open-source code and put it on GitHub. Not a degree. Not a certification. Working code that demonstrates you can build real things. That's the proof of understanding that matters — and AI can help you build it faster, but it can't build your understanding for you.

The barrier to "enough" has never been lower.

The Real Question

Nobody is arguing that you need to handwrite a binary search tree to build a startup. The question is whether understanding how computers work gives you something AI can't give you.

It does. It gives you a way of thinking — systematic, compositional, comfortable with abstraction — that transfers everywhere. Into how you structure an argument. How you debug a process. How you break down a complex problem into pieces you can actually solve.

That's not a narrow technical skill. That's a thinking tool.

Summary and Key Takeaways

✅ AI tools require judgment to use well — judgment built from actually coding
✅ The mental model from learning to code transfers far beyond programming
✅ AI raised the floor on junior work, making deeper understanding more valuable, not less
✅ You don't need to become a professional developer to benefit from learning
✅ The threshold of "enough" is low and getting lower with AI as a learning tool

References

Will AI Replace Coders? — Interview with Anders Hejlsberg, creator of C# and TypeScript, Technical Fellow at Microsoft

Final Thought

Learning to code in an age of AI assistants isn't about competing with the AI. It's about becoming someone who can use it well — and understand what it gets wrong.

The people who give up on learning because AI can do it are making the same mistake as someone who stops learning to drive because taxis exist. Maybe you won't drive every day. But knowing how gives you something that depending on taxis never will.

Learn to code. It's still worth it.