TL;DR - Skill is a big library of recognized patterns. You build it with volume plus a feedback loop. In the AI era it's worth more, not less.
Getting really good at coding comes down to one thing: pattern recognition.
The chess proof. Grandmasters don't calculate more, and they don't have alien IQs. They recognize patterns - tens of thousands of them. Show a master a real game position for five seconds and they'll rebuild it almost perfectly. Show them a random board with no patterns in it, and they do no better than a beginner. It's not memory. It's not IQ. It's a pattern library.
Same thing in coding. A junior reads code line by line and has to think about everything. With enough reps, you look at a function and just see it: the N+1 query, the race condition, the memory leak. You're not reading faster - you're recognizing chunks.
How you build the library: volume. Every rep, every bug, every project, every codebase you read adds a pattern. There's no shortcut - you can't read or watch your way to it. You have to do the reps.
But volume alone isn't enough - close the feedback loop. Coding without checking builds wrong patterns. The loop: do it, check whether it actually worked, correct. I wish I knew this earlier - I'd be much better now. And nowadays AI makes feedback cheap: after a long session, push yourself to about 70% of your limit, then use the rest to review what you built and find what could be better. Do it daily, not just when a code review happens. You'll improve much faster, I promise.
The AI angle. An LLM is itself a pattern-recognition machine. Now that AI does the syntax and the typing, your value moves up the stack: choosing the right architecture, catching when the AI is subtly wrong, knowing which pattern fits. Pattern recognition is worth more in the AI era, not less.
You're not slow because you're dumb. Your pattern library is just small. Do the volume, close the loop, and one day you look at a problem and instantly see it.