I guess I’m writing about Claude Code now.

Here’s a narrow case where I found it useful.

UI work, whether it’s front-end development or design, is something I only like in small and smaller doses. It’s fun to do every once in a while, but my heart lives with servers and systems.

This blog very intentionally only uses Astro and Tailwind. All the content is either minified images, CSV, Markdown, or JSON. No extra JS frameworks or libraries.

Repo’s private, but here’s a snippet from my package.json

  "dependencies": {
    "@astrojs/cloudflare": "^12.6.12",
    "@astrojs/mdx": "^4.3.13",
    "@astrojs/rss": "^4.0.15",
    "@astrojs/sitemap": "^3.7.0",
    "@astrojs/tailwind": "^6.0.2",
    "@tailwindcss/typography": "^0.5.15",
    "astro": "^5.18.0",
    "tailwindcss": "^3.4.17"
  }

This has made it very easy for me to do the thing I wanted to do, write in Markdown and post on the internet. I was still limited however by where I fell on the Full-Stack spectrum. I could do basic things like make a page and cards for my photos and projects that consume data for each at build time. I was able to add some pinned posts to my homepage. I even briefly had a contact form when the site was still deployed to Netlify. But I was slow to make site specific changes like required more lookups like adding proper tagging, good alt text, a search function, mobile friendly menus, a dark/light mode toggle.

The coding agent running on my Ghostty terminal does a good enough job implementing those for me. For a relatively simple use case like this of building a personal website. The basic loop of describe the overall architecture and the feature you want, pass it onto the clanker, validate the output, and then threaten repeatedly till the desired output is produced. Very similar to raising a desi child. Though unlike it, the agentic workflow has an exit loop where you do it by hand.

Best part, I can still keep shipping the minimal amount of JavaScript needed to get my webpages across the tubes. Hit F12, you can check for yourself.

But to address the elephant in the room.

Anthropic’s revenue exceeded $5 billion, but had spent over $10 billion to train and deploy its models. It’s been a super useful tool, but I do wonder how long till the bills come due and enshittification starts to happen.