Claude Code is SO good!

Claude Code blew my mind. It is SO good!
I spent about a day with Claude Code fixing up old bugs on my website that I never made time to fix. The fixes are all in my private repo, but they were all the same bugs listed in the public one here.

What blew my mind is my mental shift from coding things myself to just asking it to do them for me. 10 minutes in and I stopped writing code myself. It still takes some back and forth, and it doesn't yet get everything right, but the fact that I didn't have to write code myself was a big shift.

What Claude Code did

  • Made my website responsive. 
    My website had a whole lot responsiveness problems. CSS is a tedious little monster and so I never got myself to learn Bootstrap and other front end frameworks.
    With Claude Code, I just had to tell it in English what I wanted it to do, and it did them exactly as I wanted them. No more CSS for me!
  • Added Dark Mode.
    I've always wanted dark mode. With Claude Opus 4 it basically three shotted it, changing all files that had to be changed. This would've taken me a whole day to do.

Here's why Claude Code is so good:

  • It reads code context and creates a gist for future instances of itself.
    Windsurf and Cursor didn't do this when I tried. They need you to @mention relevant files and folders you want the AI to read. That breaks the magical experience because it's still very much on me.
    Claude Code on the other hand reads the source files and creates a CLAUDE.md file that are used for future instances of itself.

  • Straight from the terminal. 
    LLMs are text mediums. Code is a text medium. The terminal is a text medium.
    It just makes so much sense to use LLMs from there.

    I used Windsurf and Cursor for a month on their paid plan, but the separate chat window feels like a forceful addition. Those IDEs solve the problem of copying things from LLM websites and just directly editing code in place, but their integration never really worked for me.

  • Claude Code lists out what it's going to do like a check list.
    This is my favorite thing it does. Whenever I use the website version of Claude, I ask it to create the steps it will do without writing any code. With Claude Code, you get that, but with a pretty UI.


Comments

Post a Comment

Popular posts from this blog

Shake It Up | Time to leave BigTech

tl;dr: I am a senior software engineer, and I know I can easily do this work for the rest of my life. So, it's time to shake it up. - Big tech "levels" people, putting them in categories of seniority. I've reached the "terminal" level at BigTech - the Senior Software Engineer. A place where a significant number of software engineers spend their lives. I know for a fact that I can do this work at this level for the rest of my life. The tech industry has* jobs at this level. Given where I am today, a 30 year old with limited responsibilities, I have the freedom to shake it up.  Join a startup, change my role, start all over, work as a barista, work at a farm. The worst that I can do is stay where I am, that is, continue with my current team. Slightly better is to change within BigTech, but I'm pretty confident it'll be the same problems any team I switch to. So, I must shake it up. Do something where I don't know what that future will be.  I can al...
Continue reading »

Growth at BigTech

Growth at my time at BigTech meant several things to me: Become the person everyone in my working radius knows. Be responsible for a good chunk of work, and in turn get promoted for it. Learn how a big org works, and learn to be effective in them. Eventually be in the room that leaders make decisions. Based on these metrics for growth, here's what I learned you need: Visibility. Visibility Visibility. Assuming you're already a top tier engineer, a team lead, and a go-to person in your project, here's how you get visibility: Be where leaders are. Don't try to play the ladder game from a remote office. It's futile. It's like trying to climb an extra greasy ladder. You will eventually get up there, but no one will recognize nor reward you for having climbed the greasy variant. Look where leaders are looking. Don't work on things you think are important. Don't work on things your users think are important. Work on things leadership thinks are important. T...
Continue reading »

P(ai)r Programming

Over the last 2 months with Claude Pro I've reached the point where I work with AI the way I work with humans. I'm working on a fairly non-standard app - it's got a long living foreground service and interacts with the device from that service. And I'm surprised how helpful it has been for this non-standard use case. Simple tasks First, for simple tasks, I tell the AI what to do and ask it explicitly not to write code. We first review what it's going to do before letting it code. It takes us about 2 to 3 iterations, and then it codes. For most tasks, it's really good. The reviewing started as a way to work around Claude's tight token limit, but now it's just a really good tool to make sure Claude and I are on the same page. This is exactly what I do at work with my team. Every meeting is us writing high level bullets of what they will work on, and once we're on the same page, I just review their code - which usually takes them 1-2 days to come back w...
Continue reading »