How I'd Learn to Code in 2027 If I Started From Zero

Programmers are declining. Developers are growing. Both are true.

The latest US labour projections show employment for traditional computer programmers falling by around 7% over the next decade, while software developer employment grows by around 10%. That sounds contradictory until you look at what each job actually is.

Companies need fewer people whose main value is taking instructions and turning them into code. AI is extremely good at that now. What companies still badly need are people who can understand a problem, design a system, use AI intelligently, work out why something is broken and ship a working product. The World Economic Forum still lists software and application developers among the fastest-growing jobs to 2030.

So no, coding is not dead. But the definition of being good at coding has changed. I taught myself to code after leaving a job in venture capital in 2019, used it to build startups, and today run Sigmaschool, an AI-native software school. If I woke up tomorrow knowing nothing, this is exactly what I would do. (If you want the longer argument, see is coding still worth learning.)

Step 1: Ask what you want to build, not which language to learn

Most beginners start with "Python or JavaScript? TypeScript? Ruby?" That is the wrong first question. Languages are tools. You do not walk into a hardware store and announce you want to master the hammer. You go in wanting to build or fix a table, then pick the tools for the job.

If you want to build web applications, SaaS products, internal tools or full-stack apps, start with JavaScript and move to TypeScript quickly. TypeScript became the most used language on GitHub in 2025, overtaking both Python and JavaScript. If your goal is AI, machine learning, automation or data, start with Python: nearly half of new AI projects on GitHub in 2025 were written primarily in it.

But here is the part that matters most: your first language barely matters. Pick one. Do not jump from Python to JavaScript to Rust to Go because someone on Reddit said it is the future. What matters is the concepts underneath: variables, functions, loops, conditionals, objects, arrays, asynchronous code, APIs, databases, state, inputs and outputs. Once you understand those, switching languages is just syntax. More in our guide to your first programming language.

Step 2: Stop watching coding tutorials

This was the biggest mistake I made. You watch someone build something on YouTube, you type exactly what they type, everything makes sense, and after four hours you feel pretty good. Then you open a blank editor the next morning and think: where do I even start?

That is the illusion of competence, and I was stuck in it for a long time. Watching someone solve a problem is completely different from solving it yourself. So I would follow one simple rule: spend 20 to 30% of your time learning and 70% building. Learn functions, close the tutorial, write your own. Learn APIs, close the tutorial, connect one yourself. Learn databases, close the tutorial, build something that stores data.

And when it breaks, good. The moment you stare at an error wondering why it does not work is exactly where your brain starts to click. You form a hypothesis, test it, inspect the system, change something, and now it works. That feedback loop is programming, and nobody, including AI, can hand it to you. If you have been here before, read tried to learn coding and gave up.

Step 3: Use AI from day one, but in three levels

This is where my advice is very different from what I would have said five years ago. You should absolutely learn to code with AI. In Stack Overflow's latest survey, 84% of respondents use or plan to use AI in their development work, and 51% of professional developers use it daily. It is no longer an optional extra.

But beginners make one big mistake. They type "build me a Netflix clone", Claude generates 4,000 lines, it sort of runs, and they believe they are software engineers. They are not. They know how to operate Claude. So I would learn with AI in three levels, in order:

Step 4: Build one complete application

Tiny tutorials stop adding value quickly. Build something real. It does not need to be the next Facebook. Say you build an AI career adviser. You need a frontend, sign-up, a database, authentication, an API, a connection to an LLM, saved conversations, error handling, and deployment. You will learn ten times more than someone doing their 47th JavaScript tutorial, because real software forces different systems to work together.

This is where you discover something most beginners miss: software engineering is not really about code. It is about systems. The frontend talks to the backend, the backend talks to the database, the app talks to external APIs, authentication controls access, tests protect behaviour, Git tracks changes, GitHub lets you collaborate, and deployment turns a project on your laptop into something people can use. Once you understand how those pieces fit, you stop being "someone who knows JavaScript" and start becoming a software engineer.

Step 5: Learn the boring stuff

Git, GitHub, APIs, databases, HTTP, testing, debugging, logs, observability, deployment, basic system design. I find a lot of this boring too. I like building things and seeing people use them. It is tempting to skip straight to agents. Do not.

AI makes the fundamentals more valuable, not less. If Claude generates 500 lines of broken code, someone has to work out what went wrong, and Claude cannot always tell you. No serious business, say a trading brokerage, will rely on vibe-coded software to handle its errors. They need people who understand every line going into the codebase. Do not assume that because AI generated it, it is correct. Your job is now the judgement and the review. We cover the trade-off in vibe coding vs fundamentals.

Step 6: Build proof, not certificates

Your goal is to get hired, and getting hired needs evidence. Saying "I know React, I know AI" means very little when anyone can claim anything. Saying "I built this, here is the live product, here is the repo, here is how the architecture works, here are the trade-offs I chose, and here is a bug I hit and how I fixed it" tells an employer infinitely more.

Employers are trying to reduce hiring risk. Your portfolio has to answer one question: can I trust this person to solve the problem we are hiring for? I would much rather see three serious projects you deeply understand than a stack of certificates.

The 12-week roadmap I would give myself

Here is the plan, kept deliberately simple:

Code is cheap. Judgement is not.

In 2019, learning to code meant writing instructions that told the computer exactly what to do. Now computers can increasingly write those instructions themselves. Your job moves a level up: understand the problem deeply, break it down, design the system, direct the AI, inspect what it produces, fix what it gets wrong, and be accountable for the result. And ship something people actually want.

Lines of code are no longer how anyone judges a developer. Code is cheap. Judgement is not. So do not spend the next year trying to be the fastest typist of JavaScript. Learn enough code to control the machine, then learn how to build.

An honest note: you can follow this roadmap alone, and some people do. Where most self-learners stall is weeks 5 to 12, when the projects get real and nobody is reviewing their work. That is the gap our 12-week AI-Native Software Development Programme is designed to close, with mentors reviewing what you ship. If you want to try the first lessons and see whether you enjoy this before committing to anything, start the free trial.