Door #3 · The one I use
Claude Code (the terminal one)
Same Claude brain. No app, no website. You type in a black box and Claude reads your files, writes new ones, runs commands. That's it. It's the one I use every day — because there's nothing in the way.
Why I only use this one
The website and the desktop app are great. But they're polite. You ask, they reply, you copy the answer somewhere, you go again. Every loop has friction.
In the terminal, Claude doesn't tell me to change a file — it changes it. It doesn't tell me to run a test — it runs it and reads the result. The whole thing collapses into: I say what I want, Claude makes it happen.
That's what I mean by "pure brainpower." The medium disappears. It's just thinking, then doing.
What it actually looks like
It looks like this. (Yes, that's it. A typing-and-thinking screen.)
$ claude
> build me a one-page site that explains the three Claudes,
with a mobile-friendly vote for the family
● I'll scaffold an Astro + Cloudflare Pages site...
● Created package.json, astro.config.mjs, layouts...
● Built 7 pages. Wrote the vote API + Svelte form.
● npm run build → success.
Want me to deploy it now?What it's great at
- Building real software. Sites, scripts, automations, games. End-to-end.
- Touching your computer for real. Reads and edits files. Runs commands. Installs things.
- Long, complex tasks. It plans, executes, checks its own work, and reports back.
- Parallel work. I can have it building three things in three folders at once.
Where it's a worse fit
- The first 10 minutes look scary. A black window is a vibe shift if you're used to apps.
- It's a "doer" tool. If you just want to chat, the website or app is faster.
- You need to give it permission to do real stuff. Worth it — but it's not magic, you're in the loop.