Notes on Claude

Claude is on my mind, and has entered the majority of my conversations. Here are some notes on what I’ve been seeing, changing my opinion on, and recommending.

I’ve changed my mind about worktrees

Last summer, when I first shared about the power of Claude Code, I mentioned that I never use git worktrees, because I do a lot of careful review of code and interaction.

That’s no longer the case. I’m fully into worktrees, running multiple Claudes. Reviewing code looks very different nowadays. Most of my work is prototype-y and experimental, and the speed of development is blowing my mind.

I’m loving Claude Cowork

Think of it as the web chat interface you’re used to with ChatGPT or Claude or Gemini, but with access to a bunch of tools: your calendar, file system, Chrome, apps such as Granola.

Because of all this tool access, it can get a lot of tasks done on its own. I’ve had it find call transcripts to help me prep for follow-up calls, help me understand research papers by drawing very helpful diagrams (the web UI didn’t do this well), add to and directly update the GitHub repo where I organize my Claude projects and skills, shop for faux house plants, and more.

Cowork is hard to find, and pretty immature in some ways (I can’t get some of the connectors to work, so I’m using some hacky workarounds), but I’m very obsessed with it.

The course I keep recommending

I still love this deeplearning.ai course, and recommend it all the time for people getting started with Claude Code. It’s best for people who need to get past whatever inertia is keeping them away. It’s so easy to follow along, I love the style, and I find it a lot more engaging (and educational) than Anthropic’s own course.

A podcast I actually enjoyed

I enjoyed this Lenny’s Newsletter podcast with the creator of Claude Code, Boris Cherny.

My favorite part is the sheer joy that came through, about how it’s possible to build and execute now. My least favorite part is this addiction we now all have to babysitting our various Claude Code sessions at all hours of the night and day. It stinks that moving faster does not result in people working less.

I’m allergic to most tech podcasts, unfortunately, but this one is great.

Claude’s Constitution

Finally, I really appreciate that Anthropic has released Claude’s Constitution, its guiding principles. It discusses and instructs Anthropic’s models on how to think about tradeoffs between being ethical, helpful, safe, and so on. I find the casual style in which it discusses these very consequential decisions fascinating.

That’s it. I promise not to mention Claude for at least three months.