Is It Still Worth Learning to Code? A Tech Founder's Take

Start with what the machine is genuinely good at

If you skip this part, everything else is just comforting yourself. AI is better at code than most people want to admit. It writes a thousand lines in seconds. It translates a program from one approach to another like translating a book. It builds the screens and buttons you tap. It finds bugs and fixes them. Anything boring that you would otherwise do the same way a thousand times, it does instantly.

I tested it properly. I asked it to build me a dating app: swiping, profiles, a database for user data, everything. It gave me a working app within an hour. (The only thing it could not do was find me a date.) That was a real prototype, and genuinely impressive.

But look at what actually happened. I decided it should be a dating app and not a food delivery app. I decided swiping was the right interaction. Every choice was mine. The machine did the typing. I did the thinking, the choosing and the deciding.

Four things AI still gets wrong

"AI has limits" is a useless sentence. Let me be specific.

Almost right is the dangerous kind of wrong

Completely wrong is easy: you spot it in seconds. Almost right passes every test, ships to a real customer, and breaks months later, when you have to dig through code nobody remembers writing.

So there are two ways to be in trouble. If you do not use these tools, you are slow, and 84% of the people you compete with are not. If you take whatever comes out and ship it unchecked, you are fast in the wrong direction, and you are a commodity, because anyone can prompt and paste. One group loses slowly. The other loses quickly. More on that balance in vibe coding vs fundamentals.

Treat AI like a very smart new hire

These tools all learn the same way: they read an enormous amount of what humans have written and get very good at guessing what comes next. So think of AI as someone brilliant you hired last week. Extremely fast, knows a shocking amount, works at 3am, never complains. Has no idea what your company is trying to do, has never spoken to a customer, and will happily do the wrong thing beautifully if you explain it badly.

What do you do with someone like that? Give clear instructions. Check the work before it goes out. And carry the blame, because you are the one who let it out of the door. You cannot blame the AI.

This also demystifies prompting. People treat the perfect prompt like a magic spell. A prompt is just instructions. Giving clear instructions is the same skill as managing a junior. If you cannot explain what you want, why, and what done well looks like, no template will save you. And your smart new hire cannot sit in the meeting, notice that two departments want opposite things, or tell that the client said yes but meant no. The machine handles the code. We handle the understanding.

The job was never typing

Building software was never really about writing code. Code was just the visible part. The real work is three things: working out what someone actually needs (almost never what they said in the meeting), working with people (designers, the business side, the client who changes their mind on Friday afternoon), and making hard calls and owning them, because real people will use this and something will go wrong.

The best engineer I have worked with was not the fastest typist. He thought the hardest, went the deepest and understood the system as a whole. Hand him a messy, half-explained problem and he would turn it into something clear enough to build properly. That skill, taking a mess and making it clear, is worth more now, because AI needs clarity before it can do anything useful. Hand it a mess and you get a mess back, faster and with nicer styling.

If anyone can build, why do we still need engineers?

This part is real and I will not downplay it. A designer can now build a working app without any engineer. A marketer can analyse data without waiting weeks. That gate is gone, which is great for business. It also means some of what engineers were paid for a few years ago is gone. So do engineers lose? No, for three reasons.

One: engineers know how it works inside. Anyone can ask the machine a question. Very few know how it was built, why it fails the way it does, or what private data it might leak. When AI is inside almost every product, knowing how the engine works is not a bonus. It is the job.

Two: working once is not working properly. Something that works once has to run on your laptop one time. Something that works properly has to handle tens of thousands of requests from different people at once, survive a part breaking at 3am, never leak anyone's private data, and still be fixable in two hours by someone who did not build it. The distance between those two is enormous, and it is basically the whole job now.

Three: engineers are the ones making AI better. Someone has to improve, maintain, speed up and cheapen these tools, and build the products that put them in your hands. The prediction from GitHub's boss is not true yet, and the people making it true are engineers.

The easy part is free. The hard part did not get easier.

Making something that works is now free. Anyone can do it in an afternoon with no training. The hard part did not get any easier: knowing what to build, knowing when the answer on your screen is wrong, getting five people to agree, building something that still stands in two years. Nobody has automated that, and nobody is close. All the value has moved into the hard parts, and most people are still training for the easy part. That is a training problem, so here is how to fix it:

Use AI as a teammate, and stay a beginner

Two more. Use AI like a teammate, not a vending machine: learn how the tools work underneath, and talk the problem through to understand it before asking for an answer, the way you would think out loud with a colleague. And get comfortable being a beginner again. Tools used to change in months; now they change in weeks. Whatever you learn this year will be dated in three. The real skill is learning new things quickly, and you only build it by deliberately learning uncomfortable things.

When everyone can build a little, the person who really knows what is going on becomes the one who leads. I am not even sure "programmer" is the right word any more. The job is deciding what is worth building, getting people together to build it well, and leading both the humans and the machines. The future belongs to whoever thinks further ahead, thinks harder, learns fastest and works best with others.

So, is it still worth learning to code? Yes, if you learn the hard parts, not just the easy part AI already gives away. That is exactly how our AI-Native Software Development Programme is designed: fundamentals first, real systems, AI as a teammate, and mentors who make you explain what you built. If you want to try it, start the free trial. Or take this week's challenge: pick one real problem in your life and build something that works by Sunday, even if it looks ugly. For the written long-form version of this argument, see is coding still worth learning?