Previously on my blog, I wrote about vibe coding and how I was experimenting with Google CLI, the free agentic AI thing that runs in your command line. I talked about how cool it was, but also how I was too cheap to try anything that cost money. After repeatedly hearing about how good Claude Code was, I decided to scrounge up the last few dollars I had in my vast money bin and spent $20 on Claude Pro. So was it worth it? Heck yeah it was! Claude Code is my new best friend!

What is Claude Code?

Claude Code is Anthropic’s official CLI tool that brings Claude’s coding capabilities directly to your terminal. It’s a lot like the Google Gemini CLI that I talked about before, but it actually has brains and doesn’t get stuck in an infinite loop constantly for no reason. Previously, when I tried using the Gemini CLI, it would mostly hit rate limits, switch to the flash version of the model, and generally be bad at doing things. Even with the degraded quality, I still found some use for it, though. But I knew that if I wanted to be a true Vibe Coder, I’d have to upgrade.

Claude Code uses Claude Sonnet 4 which is apparently really good at code. So far I’ve used it to:

Code:

• I added automated testing with playwright to Animenano
• Optimized Animenano RSS feed parsing by caching etag/last-modified headers
• Wrote a script to sync the remote Animenano Cloudflare D1 database for testing locally
• Added some tools to my LiveKit AI agent that I was messing around with, like a Japanese phrase guessing game
• Vibe coded search on my blog on a plane! (I actually made a video for this!)
• Set up staticman for comments on my blog!
• Changed the comment form on my blog from a hard reload to an AJAX update!
• Wrote a script to scrape comments from my blog posts on archive.org to replenish my missing comments!

Other:

• Documentation for my various side projects, instead of leaving the readme blank
• Set up wireguard on my opnsense, Claude Code helped me debug why I couldn’t access the network until it worked!
• Added documentation on how I set up mdns and nginx on a home server to serve stuff on the network, and added some more hosts

The “Other” category is honestly a really cool unlock for me personally. I have a bunch of home servers that do random stuff like run cron jobs of scripts for automation and other really random things. I usually just write one-off implementations based on quick searches of how to do things (like setting up mdns and nginx to redirect example.local domains to 10.0.0.111:8000 for example). Inevitably, I’ll forget how I set something up (like when I set up a launchd daemon thing) and it becomes this tech debt where it becomes really difficult to fix or change anything if something breaks or I need to set it up again a different way.

With Claude Code, I can just have it read the files and figure out how things work, then write a README.md that describes in plain language how it’s set up, how to use it, and how to modify it. Claude Code is incredible for side projects and these sorts of things where you naturally don’t put the same amount of rigor into documentation as you would for professional work. It takes the boring parts of fun projects and automates them, so you have more time to start even more side projects that may or may not ever be finished!

There’s a Plan For HOW MUCH?

I’ve been seeing a lot of posts online about how people are YOLOing their wallets and buying the $200 a month Claude Max plan, which includes 20x the amount of requests that you can make with the Pro plan (which I’m currently on). I suppose that if I was using Claude Code for my job, and it was on a super large codebase, that might make some sense. But honestly, it seems a bit excessive. For an experienced software engineer, coding is probably like 10% of what you actually do, and the rest is reading code and planning what to do. I guess Claude can help with that too, but it just seems like the guys who are running 8 subagents simultaneously and racking up the Claude Code tokens could really just learn to git gud at software engineering. Or as the kids would say, Skill Issue! They’re obviously just karma farming because who actually does that!?

I guess one additional bonus of that ridiculous MAX plan is that you have access to the Opus 4 model, which may or may not be named after that penguin from the 80s. I already feel like I can accomplish pretty much whatever using Sonnet, but I guess that if I hit too many roadblocks, it might be interesting to at least try Opus.

I do think I made a pretty good decision paying for the $20 plan as I rarely hit the rate limit anyway, and I’ve been using this tool called ccusage to see how much I’d be paying if I was using the pay-as-you-go style of token based API usage.

This image doesn’t completely reflect my usage because it’s divided between like 3 different computers, and the tool only shows usage for the computer it’s running on. And I’m too lazy to go run it on all of them right now to see how much I actually used. On top of that, I think the Claude Code Github Actions stuff uses your Pro Plan quota, so that’s not showing up in the numbers either.

I plan on continuing to use Claude Code and paying for the subscription for now. I started it on July 1st and I’ve already gotten my money’s worth for the month. To be honest, I have a suspicion based on the token usage image above that this price is heavily subsidized, kinda like how Uber and Lyft rides used to be super cheap. Though with those things, you’d have to pay a person, and with Claude, you just need to pay to run a server. Hopefully servers get cheaper or more efficient, but even so, it’s hard to believe that this much utility can come at such a low price. Maybe I should be more bullish about AI.

I should probably report back in a month or so to see if I run out of things to use CC on, or if I just keep using it as frequently as I am now. For now though, I’m totally bought into the hype. Claude Code forever!

BTW

Also, I’m still using this AI stuff to help me write my blog. None of the actual words were written using AI, but I did use it to help me be extremely lazy. For example, the image I made at the top of this blog (which was sorta generated with AI but also hand edited) was plopped into the root directory of my Jekyll repo. I then told AI to put it in the right directory, and make a link and a caption to it. I did the same with the screenshot of the ccusage tool, and I also told the AI to resize it since it was kinda big. It used a command line tool and got it to 150kb from 650kb. Interestingly, I tried asking it to add some alt text to the images in this post for accessibility. It did an okay job at first but I kept nudging it, and now the images are a lot more accessible. I also had it add some links within the post, which I could totally do myself, but why should I do it if AI can do it for me?

To me, that’s the whole point of AI. I enjoy writing blog posts and getting my thoughts out there, but wrangling images and formatting captions brings me no joy; it’s just friction. AI is helping me be creative and express myself, without taking away the soul of the process of creation, which is the way it should be. I’m sure I’ll think of more ways to use AI, but for now this is getting me really excited about the possibilities!