In LLM coding, people will spend weeks learning frameworks and tooling, and spend basically zero time learning how to talk to the thing that’s actually writing the code.

I’m pretty sure I’m the #1 power user of coding AI tools at my company (Claude Code, Codex, etc). People asked me to showcase how I use them. As I’m planning it, I notice the main idea I want to share is: using these tools is a skill. Like sports. You don’t get good by copying someone else’s “perfect routine”.

The important part is reps: time + effort (and yeah, sometimes spending money on tokens). If it’s handed to you, you won’t level up. Or even find it useful at all. And yes, it may even slow you down at first. Learning has to be a bit hard. Sometimes slow. That’s normal.

It’s like coaching a swimming team by showing slides of the physics of swimming. Not useless, but you still need to get into the pool (A LOT!).

What actually works: sit down, write your own prompts, test what works/doesn’t, iterate. Build your own little tools/workflows to access LLMs.

Break stuff.

Fix stuff.

Repeat.

Copy-pasting prompts/tools is almost worthless long-term. Steal concepts, not templates. The Ralph Wiggum technique is just another concept to steal. Write your own prompt, run each step of the loop manually at first to see how it fails or succeeds. Do NOT just install a claude-code plugin and assume you’ve learned.

Same with everything else. If you see a new technique, don’t just copy the prompt, take the idea and recreate it in your own way.

With LLMs, you must learn the failure modes, where it misunderstands you, where you’re vague, where structure matters. Biggest unlock: use LLMs a lot. Even when you “don’t need to”. This is already hard to do with any tools. Even harder with LLMs because they are non-deterministic, and get better every few weeks. So you always need to be trying stuff.

Try it on big tasks. Ask it to plan first. Debate approaches. Compare pros/cons. Then implement. Also on small tasks. Give it read-only access to the database and ask it questions. Use it to explore new codebases. Use it for everything.

YOU DO THE THINKING! The LLM the work.

tl;dr

If you want to get better at using LLMs in general, just use them. If you don’t know what to use them for, then get more ambitious and creative.