I’m looking forward to this semester’s vibe coding workshop for librarians. I was notified that I was accepted into the program yesterday. It should be fun! I’m told we’ll be using Claude Code, so I set this up on my laptop last night.

First of all, I don’t particularly trust Claude Code. Allowing an agent to autonomously run code of its choosing is rather terrifying. My response has been to sandbox it, even beyond Claude Code’s default precautions. I’ve created a VM on my laptop using virt-manager that I will use exclusively for Claude Code, and nothing else. At worst this is prudent but unnecessary.

Also, I’m amused by Claude Code’s CLI. It is the most GUI-fied CLI I have ever encountered. I’m pretty sure that if you showed this to someone in the 90’s they would definitely call it a GUI. Anyhow, I’m not complaining. It’s better than using vscode, which is the alternative.

Lastly, Claude Code did rather well at the tasks I gave it as an initial test. Maybe the underlying models have improved since I was playing with Aider, or maybe Claude’s agentic approach is simply more powerful than Aider’s pair-programming strategy. Anyhow, the progress in this space is very palpable.