I shifted gears from using Claude (Sonnet 4.6) with Hermes on PC (via Telegram) to using Claude Code in Terminal on my MacBook Air. It did not go as planned.
The first thing I did was ask it to build me a to-do app. It built me a to-do app. Then I asked it to make the to-do app into a note-taking app. It built me a note-taking app. Then I asked it to integrate the two. It sent me a bill.
I’m kidding. But only slightly.
The real lesson here is that Claude Code is a tool for people who already know what they’re doing. It accelerates competent people. It does not replace competence. I spent three hours having it build things I didn’t understand, then spent five hours trying to figure out why nothing worked.
Here’s what I should have done:
- Understand the stack before asking it to build anything
- Use it to write the parts I understand but that are tedious
- Read every line of code it produces before running it
Live and learn. Next post: building something that actually works.