How I used Claude Code for Ops today

Bonus: The magazine entry he helped me get publish!

Epistemic status: I actually use Claude for Ops and it’s great. Also, the second part is mostly a joke.

According to Anthropic, Claude Code is “Deep coding at terminal velocity”. It’s true, but it’s also so much more! In my hands, it has become Operations at terminal velocity.

Here are two ways I’ve been using Claude Code today:

Keeping the Inkhaven calendar up to date

I’m responsible for keeping the Inkhaven activities calendar up to date. Activities are announced in a free form format in a Slack channel, get moved around the campus, change day, change time, get cancelled, become recurring, add requirements, etc. There are so many special cases that it makes it basically impossible to make an automated system to process the messages. Before Claude Code, my only solutions would have been either requiring people running activities to add them manually to the calendar and keep them updated, or for me to do a lot of manual work.

Now, I just ask Claude “update the calendar”, and he fetches the whole channel history, fetched the current state of the calendar, and presents me with his plan of which calendar events to update or create. Often it’s just correct, and he then adds them to the calendar. Then, I ask him “draft the announcement” and he makes a compact summary of the event of the day, which I slightly edit and then send to inform the other residents of the planned activities.

Sometime, his output is not perfect. He forgets an important info in the event title, he gets confused about a date, he does not do a great job at formatting the announcement. When this happens, I give him feedback, and he updates his SKILL.md file to make it right on the first try next time.

Now, he’s just right so often. I do all this work in 1 minute instead of 10–15 minutes.

Claude Code working and my SKILL.md

Analyzing the Inkhaven publication record

I’ve also used Claude Code to do data analysis. Usually, writing Python is tedious: you have to set up your environment, install packages, remember the specific pandas methods for accessing the right slice of your data. Instead, I do Vibe analysis: I ask Claude a question about the data, and he makes a one-off script specifically to get this answer. Claude is also great at building visualizations using Matplotlib, and he does it so fast that I can ask it to present the data using lots of different lenses, and get a general feel for the dataset quickly.

I used Claude Code to do the data analysis of the Manifold.love profiles for this post on Monday:

I also used it today to check how the number of words in residents’ essays were evolving. The finding were concerning. You can find more about this story in my publication in a prestigious publication, entirely produced in one afternoon at Inkhaven, The LOOP.

Good thing Claude Code was here to warn us of the danger.

Conclusion

Claude Code is great. He’s my best coworker.

How Claude drew itself as “Claude Ops”