A Week Of Claude Code
This week at work I had the chance to spend some time working with Claude Code and I loved it!
I’ve already processed a lot of information about Claude Code, so I started off feeling pretty comfortable with using it. I am so aware there is a lot about Claude Code out there, so my remarks will really only focus on my experience. Additionally, if there is one thing that is true, its that the AI industry moves fast. Who knows… will this article be completely irrelevant in 1 month or in 1 year? So I’ll try and balance the constant and the trendy.
Context: I’ve used Claude code for a variety of different tasks, but I primarily tested it with a brand new project and an existing + extensive + open source project.
Did you know that Claude was possibly named after the Mathematician Claude Shannon?
What did surprise me
I knew Claude Code was run in the terminal, but I didn’t realize it mattered where it run. I love that approach to context control. It felt like a very practical and intuitive design choice.
The Claude.md file was more helpful then I gave it credit for helping Claude process a large codebase. It really simplified the context gathering later on. It is less helpful for smaller projects you might be starting from scratch.
The initial tools were limited. I didn’t realize it wouldn’t have access to the web right off the bat. I didn’t experiment with other tools, preferring to keep it quite bare-bones.
Other Notes
Claude code is so fun to work with. I seriously went so much farther then I would have normally. Getting into a flow state was significantly easier.
Reading text isn’t fun. Andrej Karpathy in his talk Software Is Changing (Again) emphasizes speeding up the cycle between generation and verification. Like Andrej said “reading text is effortful and not fun.” Reading Claude’s diffs was fun only because I was excited about using Claude. Sometimes the text was quite extensive as the diffs grew. My normal coding practice involves small incremental diffs, so this could feel quite foreign and frustrating.
I love that it is platform agnostic. It makes it so easy to work with it regardless of your IDE. This resolves some serious pain points and makes it a lot easier for teams to incorporate without hiccups.
I loved using with git! The amount of times I need to look up basic git commands that I only use semi often is frankly embarrassing, so this felt like a godsend. Using it to synthesize information collected from the gitlog was so easy and so helpful when I worked on the open source project.
I felt most comfortable using claude if I had a good idea of what I wanted. Blindly accepting suggestions is just not my m.o. I know it will break eventually and I know the frustration that will ensue if i know what is going on, let alone if I have no idea. I am not a masochist.