How to Git Gud at Claude Code (a loose guide)

I’ve been working with Claude Code since Day 1 and AI coding tools before then. What started out as a sort of fun weird thing has begun to rapidly evolve into a more proper “discipline”. There are people that are better at prompting and understand how to “work” with AI than others — and notably this was not always the case.

A few short years ago everyone was just kind of firing in the dark, and the difference between someone who was “good at prompting” vs. “bad at prompting” was mostly a delta that could be rapidly closed after reading some key tweet or whatever.

However now, these tools have matured, and with them the workflows to drive value from them. I’m being a bit of an asshole here now but, knowing what I know now about working with AI, if someone tells me they couldn’t get an AI to do some task X, I largely assume it’s a skill issue (sorry!).

The idea of prompt engineering/context engineering/etc. is evolving, as I said above, into an actual engineering discipline. The idea that, for any given eng discipline, that you can waltz in and directly extract value based on only passing familiarity with that field is silly, and AI is becoming the same way. It is a tool that you have to know how to be good at using.

To that end, I wanted to capture in this post (and also have it live as a sort of ongoing dumping ground), of things that I’ve read or watched that I think capture what this emerging discipline is “about”.

If you feel behind in AI-driven tooling, consider this also a small syllabus to get up to speed on how to Get Things Done with AI Tools.

How I use Claude Code (+ my best tips)

This one I think is a good general dive into “Claude Code best practice” — it hits all the high notes of what I would consider idiomatic use and is a good guide to effectively using and getting started with the tool outside of reading Anthropic’s docs (though you should also read Anthropic’s docs).

AI Blindspots

https://ezyang.github.io/ai-blindspots/

I’ve referenced this blog in other posts I’ve written so I’m just surfacing it again here more formally. A big part of getting useful output from these models is starting to gain an intuitive sense of not only what they are good at doing, but also what they are bad at doing.

The best way to is to Just Make Stuff, but this series of posts goes below the surface of that to illuminate things to add to your “probably bad at this” list.

Evals FAQ

https://hamel.dev/blog/posts/evals-faq/

One thing about know if an AI is actually good at something is having a way to think about and run evals. This post is a great introduction to the idea of how you actually quantify AI output outside of just “vibes”.

Agentic Coding: The Future of Software Development with Agents

This video is a good overview into more “advanced” Claude Code use. It’s one of the things I thinking about when I think of context and prompt engineering as actual engineering.

Advanced Context Engineering for Agents

I think this video is great. Circa when I’m writing this — Sept. 2025 — I think this does a great job at representing the SOTA for working with tools like Claude Code. It dives into how to think about context and tools like subagents holistically as part of a development effort. Highly recommend this!

I know it’s not a lot for now but wanted to just create a small space for high-signal posts that I think can help people. Thanks for reading!