How to learn to code: what works and what only feels like it

The main trap

Watching someone write code is pleasant. It all makes sense, it's all logical, and you get the feeling you've understood it.

Then you sit down to write yourself and can't start from a blank file. That isn't a memory failure. It's the gap between recognition and ability: you learned to follow someone else's thinking, not to think yourself.

Only one thing cures it: writing code without prompts.

The one-third rule

For every hour of watching or reading, two hours of independent work. Not repeating after the author — solving your own problems.

The ratio can shift, but the principle holds: there must be more practice than theory. Most self-taught learners do the reverse and stall for months.

Studying so it sticks

Understand rather than copy. Took a solution from a forum? Walk through it line by line. Didn't understand it? Don't paste it.

Break things deliberately. Take working code, remove a line, change a value, see what fails. That teaches structure better than ten articles.

Explain out loud. Narrate how the thing you wrote works. Where you stumble is where you didn't understand.

Come back to your code a week later. If you can't follow your own work, you were writing mechanically.

How much to practise

Regularity beats volume. An hour five days a week beats six hours on Sunday, for two reasons:

Forgetting. A week's gap loses more than a single session gains.
Warming up. The first twenty minutes of any session go on getting back into context. With short frequent sessions that loss is smaller.

The minimum at which there's movement is about an hour a day. Less is treading water.

When you're stuck

The half-hour rule: thirty minutes on the same thing with no progress means change the approach.

Write the problem down as though asking a person. Half of them resolve at this step.
Break it apart: what exactly fails, what you expected, what you got.
Ask people.

Spending a week on one error out of pride is a classic beginner's mistake. At work that reads not as persistence but as an inability to ask for help in time.

On AI assistants

An assistant saves time when you understand what it produced. It harms you when you paste code you don't understand: the feeling of progress is there, the skill isn't.

A working rule: ask for an explanation, not a finished answer. And verify — confident wrong answers are common.

What self-study lacks

Feedback. You don't know whether your code is dreadful or fine, because there's nothing to compare it to.

So it's worth working with people early: an open source project, a team, anywhere someone will read your code and say what's wrong. One review through someone else's eyes replaces a week of wandering alone.