All video posts

Learning to code · AI · Career

Why Learn to Code If AI Can Already Write Code?

Why would anyone spend six months to a year learning to code when AI can build an entire application in minutes? It is a fair question, and I get asked it all the time. My answer: AI being able to write code does not mean AI can reliably write good software, and that gap is where your career lives.

Deric YeeDeric Yee 16 November 2026 9 min read 13 min video

Prefer to read? The full article is below. Watch on YouTube.

The short version

  1. 1Writing code is only one part of building software. Someone still has to design how it works, secure it, and fix it when it breaks.
  2. 2AI will replace some developers: the ones whose whole job is taking instructions and writing simple scripts.
  3. 3AI can be confidently wrong, so you need enough understanding to audit its work. Use it as a guide, not a crutch.
  4. 4The bar for junior developers is higher than it was five years ago. Update your map instead of preparing for an old job description.
  5. 5Learn fundamentals, learn to use AI tools properly, and build real projects people actually use.

Generating code is not the same as building software

Today you can open Claude Code, Codex or Cursor, describe the application you want and watch AI build it. You do not even need to know how to code, and it is only getting better. So if you are planning to spend the next six months memorising JavaScript syntax and working out how to centre a button, I genuinely think you might be wasting a lot of your time.

But there is a very big difference between generating code and building good software. I taught myself to code in 2019 after leaving venture capital, and I now run Sigmaschool, so I have watched this industry change up close. The people who understand that difference will have a huge advantage from here on.

Imagine asking AI to build you a banking app, something like Maybank or CIMB. Users should sign up, log in, check balances and transfer money. Within ten minutes you get something beautiful: a login page, a dashboard, a database, buttons that work. You click around, move some money, everything seems fine, so you launch it.

Then one day a customer tries to transfer $5,000 and, because of some bug in your application, $50,000 goes out instead. What do you tell your customers? That ChatGPT wrote the code so it is not your problem? Of course not.

Would you board a plane flown by autopilot alone?

Think about a pilot who tells you he knows how to use autopilot but was never professionally trained to fly. Autopilot probably handles 95% of the flight. Would you still sit on that plane? Or a doctor who says ChatGPT trained him and he can answer all your questions, but he never went to medical school. Would you let him operate on you?

Software is the same for anything serious. Someone still has to work out how the whole application should behave. What happens when two people transfer money at the same time? What happens if the internet drops halfway through a transaction? How do you stop one person accessing another person's account? When something breaks, how do you find the problem and fix it?

Those are software engineering problems. AI can absolutely help solve them, but if you do not know the problems exist, how will you know what to ask AI to do? How do you put guardrails in place, and how do you know the fix did not break something else? That is the difference between someone who can generate code and someone who understands what it takes to build software.

“Just because AI is able to write code, that doesn't mean AI is able to reliably write good software.”

Deric Yee, in the video

Which developers will AI replace?

Whenever someone tells me AI is going to replace software developers, my first question is simple: which developers? It will certainly replace some roles. If your entire job is taking instructions, writing simple code and scripts and passing it on to someone else, then my friend, you are in a bit of trouble, because that work is very easy to automate.

But look at the companies building these tools. Back in 2025, Microsoft CEO Satya Nadella said around 20 to 30% of the code in some Microsoft projects was already written by AI. That is significant, yet Microsoft did not suddenly stop needing software engineers. Someone still decides what to build, how systems fit together, and whether the code is safe and reliable enough to use.

Stack Overflow's 2026 survey found coding assistants and agents had become the most common AI use case among respondents, at 66%, yet almost half said they only trust AI's answers when they can easily check them. Developers are using AI more than ever, and they still need to understand what it is doing. We are moving from paying developers to write code towards paying engineers to solve problems using code and AI. I go further into this in will AI replace developers.

AI can be wrong, very confidently

If you think I am exaggerating the danger of relying too much on AI, here is what happened in 2025. An entrepreneur named Jason Lemkin was building an app on Replit, an AI coding platform, with an AI agent. At one point he specifically told the agent to stop making changes. The agent went ahead and deleted his live database anyway, with data on more than 1,200 executives. When he asked what happened, it told him the data could not be recovered. It turned out recovery was possible.

To Replit's credit, they acknowledged the problem and added safeguards. But the lesson stands: AI can be extremely powerful and still make mistakes so confidently that you would trust it if you did not audit its work. Sometimes you do not realise anything is wrong until the damage is done.

Now imagine that was not an experiment but a medical tool, a banking system, a brokerage or a company processing thousands of payments a day. That is why databases, security, backups, guardrails and deployment still matter. AI can help you with all of them, but blindly following its instructions will not make you productive. Use AI as a guide, not a crutch. I wrote more about this in why you should not become dependent on AI.

The job market has changed, so update your map

I do not want to pretend everything is fine. AI has made it harder for some people to land their first tech job. In August 2026, Stanford researchers published updated findings showing employment among workers aged 22 to 25 in highly AI-exposed occupations was around 19% below where it would be if it had kept pace with less exposed jobs. The researchers were clear that other factors could be involved, and it does not mean AI directly destroyed 19% of jobs. But something is changing.

Companies used to hire juniors for simple tasks: a small feature, a couple of bugs, a website update. That gave juniors time to learn on the job. Today an experienced developer can hand much of that grunt work to AI, which reduces the need for juniors doing pure coding tasks. So the bar for getting hired is higher.

The answer is not to stop learning, and it is not to go all in on AI either. It is to stop preparing for the developer job that existed five years ago. You need to become someone who can build and maintain useful software, not someone who has merely completed a coding course.

What I would learn if I started from zero

If I were starting again heading into 2027, this is how I would approach it.

01

Learn the fundamentals, not the syntax

Understand how software actually works. What happens behind the screen when you open a website? What are the front end and back end, and how do they talk? What is an API, where is data stored, and how does login keep one user out of another's account? AI can tutor you through all of this step by step.

02

Learn to use AI tools properly

Not just asking ChatGPT to explain code. Use Claude Code, Codex and Cursor to build software: give clear instructions, review what they generate, spot and fix mistakes, and test your application. Looking good and having buttons that work does not mean the software is good.

03

Build real projects people use

Skip the calculator and to-do app copied from YouTube. Build a booking system for a local business, a job-application helper for students, or an AI agent that organises applications for a recruiter. Then deploy it, secure it, handle errors and improve it when it breaks.

Projects teach you the job, not just the portfolio

Building these things is not only about having something to show. The experience of deploying, securing and fixing real software is what actually turns you into a software developer, and it gives you real stories to tell when you apply for jobs or pitch for projects. For ideas, see coding projects for beginners.

This is the core reason we teach software engineering with AI at Sigmaschool. I do not think it makes sense to teach coding the way we did five years ago. We look for the sweet spot: we do not want students skipping fundamentals and going straight into blind vibe coding, and we do not want them stuck writing syntax from scratch with AI banned either. Students build every week from the first week: front end, back end, databases, authentication, APIs and deployment, and then AI-powered applications.

My goal is not to teach people to write code faster than AI. No human will win that race. My goal is to train people to understand software well enough to use AI to build useful things. Whether you learn with us, at university or on your own at home, the principle is the same: learn how to build software, not just how to write code.

Yes, learn it, but for the right reason

When I taught myself to code, being able to write software was a massive advantage. I could build things most people could not. AI has levelled that playing field, and honestly, I think that is a good thing. More people can turn ideas into products, more businesses can build their own software, and developers spend less time on repetitive code and more on interesting problems.

It also means simply knowing how to code will not make you special any more. Your value increasingly comes from knowing what to build, building it properly and reliably, and using AI to get there fast. Do that and you will be very much in demand.

So if you are thinking about learning to code, my advice is yes, absolutely, go and learn it. Just do not learn it to compete with AI, and do not vibe code and accept whatever it outputs. Learn the principles well enough to make AI work for you. If you want a structured way to do that, our AI-Native Software Development Programme puts fundamentals first and builds with AI from week one. Not sure yet? Start with the free trial.

“The future does not belong to the people who are able to type the code fastest. It belongs to people who are able to use AI to solve problems fast.”

Deric Yee, in the video

Questions people ask

  • Is it still worth learning to code if AI can write code?

    Yes, but learn it for a different reason than five years ago. AI can generate code quickly, yet it cannot reliably decide how a system should work, keep it secure, or tell you when its own output is wrong. Learning software fundamentals lets you direct AI tools, review what they produce and fix what breaks. That combination is what employers now pay for, rather than raw typing speed or memorised syntax.

  • Which developer jobs are most at risk from AI?

    Roles where the whole job is taking instructions, writing simple code or scripts and passing the work on are the easiest to automate. Roles that involve deciding what to build, designing how systems fit together, handling security and reliability, and debugging real problems are much harder to replace. AI shifts developers from writing every line towards solving problems with code and AI together.

  • What should a beginner learn first in the AI era?

    Start with how software actually works: front end and back end, APIs, databases, authentication and servers. You do not need months of syntax memorisation. Then learn to use AI coding tools such as Claude Code, Codex or Cursor properly, including reviewing and testing what they generate. Finally, build and deploy real projects that solve a problem someone actually has.

  • Is vibe coding enough to get a developer job?

    Usually not. Vibe coding can produce something that looks good and has working buttons, but that does not mean the software is secure, reliable or maintainable. When something breaks, you need to understand why and fix it without breaking something else. Employers want people who can use AI to build quickly and still take responsibility for the quality of what ships.

Your next step

Our 12-week AI-Native Software Development Programme is a live, mentor-reviewed cohort where you build and ship real products with AI, and learn to explain every line. Start with the free trial to see how we teach, then decide.

Watch next

All videos

Learn the fundamentals that pay off in any direction.
Coding plus AI, taught properly.

Whatever you end up doing, the people who understand how software and AI actually work get to direct them. Try the first lessons free, no card needed, or see the full 12-week AI-Native Software Development Programme.