Notes ➤
I am experimenting with Claude Code for the first time. I am building a tool that monitors certain pages for changes, uses a small LLM to filter them, and takes action based on that. For example, making a push request to All About Berlin when the German minimum wage changes, or sending me a daily digest of discussions about the immigration office on different social media.
So far, it needs a lot of hand-holding to produce acceptable code. The code works, but it’s far from elegant. I have to fix a lot of not-quite-bugs such as logging warnings instead of throwing exceptions.
On the bright side, I just make comments and let it make the changes, and I find that seriously cool. For example, “this method should fail hard and fast; don’t suppress errors”. This lets me focus on what I want, and not on tedious text editing. This already makes Claude worth it for me.