How I'm learning to code with AI - and building real apps along the way

For the next two or three months I'm doing something I've called my "Summer of Code": learning to program by building apps that get progressively harder. Not courses, not tutorials disconnected from reality, but real, working projects you can open in a browser or on your phone.

The starting point was a single assumption: every project has to solve an actual user problem. I'm not reinventing the wheel. I take products that already exist, already have customers and already work, and I add my own twist. Something that I'm missing, or that the people around me are missing.

The second assumption is about method: I'm doing this together with AI (specifically with Anthropic's Claude tools), but not in the sense of "the AI writes, I click". I'm learning to run the process, to plan and design first, and only then tell the machine to write code. More on why that distinction matters below.

The plan is split into five phases, from the simplest website all the way to a mobile iOS app. I'm starting with a project called Focus Space, and that's mostly what this piece is about, because it's where everything else gets forged.

Why this path in the first place

There are two ways to learn programming. The first: you take courses and solve exercises that lead to nothing you could show anyone. The second: from day one you build things that work, and you learn exactly what the next, slightly harder project demands of you.

I picked the second one. I laid out a path of five projects, where each one is a notch harder than the last and builds on what the previous one taught me. A bit like levels in a game. You don't jump straight to the boss, you pick up skills along the way.

One rule I set for myself matters here: I don't invent new products, I improve proven ones. A focus app, a water reminder app, a book club voting app, all of these already exist. But in each one something is missing for me, or it's built in a way that puts me off. That gap is exactly my starting point.

First project: Focus Space

You know the Pomodoro technique? You set a timer for 25 minutes, focus on one task for that stretch, then take a short break. Simple, effective, been around for years.

The problem is that most Pomodoro apps look either boring or childish: growing little trees, cheerful animals, pastel colours. For some people (myself and plenty of other nerds included) that just doesn't land.

So I'm building a version for science fiction fans. In Focus Space you don't "set a timer", you launch on a mission through hyperspace. You're sitting in a spaceship cockpit, stars visible through the windscreen. You pull the launch lever, the stars blur into streaks the way they do in Star Wars, and off you go. Your focus time is your mission. When you're done, the ship slows down, the engine cools, and you get your break.

There's one more nice touch. You can switch on "Do Not Disturb" mode. If you tab away to another app or browser tab for more than 10 seconds while it's running, the mission fails. The cockpit shatters and a "Mission Failed" message comes up. Sounds harsh, but that's the point: there has to be a real stake, so you're not tempted to glance at your phone.

How I build it, without a single word of jargon

The interesting part is probably how I do this, not the what. Because I'm not sitting there typing cryptic commands from morning to night.

Imagine you're building a house. The bad way: you tell the construction crew "build something nice" and come back in a month. The good way: first there's an architectural design, then a room-by-room plan, and only at the very end does someone lay bricks, following the drawings rather than improvising.

That's exactly how I work. First I write a document describing in plain language what's supposed to get built, what's in scope, and what we're deliberately not doing right now. Then, in one tool, I design how each screen should look, like architectural drawings. And only then, in a second tool, does the AI write the actual code, following those drawings instead of guessing.

That distinction is the heart of what I'm learning. AI can write code in a flash, but I'm the one who has to know what needs to be built and why. Without that, I get something that works, just not necessarily the thing I had in mind.

What's next

Focus Space is only the first of five stops. The ones after it add new building blocks: storing user data, payments, things happening live between multiple people, and eventually a real mobile app for the iPhone. Every time: a real problem, a proven product as a reference point, my twist.

I'll be writing about it as I go. If you want a look at what learning to code in 2026 actually looks like, when AI writes the code but someone still has to steer, stick around.