I, Claude (or, the upcoming feature progresses)

I spent only a couple of hours today working with Claude Sonnet 4.6 on a long-planned feature for this application. It's not exactly a blog feature, and in fact initially it will only show up in the Calendar. I'm just psyched because in a short time working with Claude, I got a full spec for the feature, which Claude then used to create a passable first draft of it. A couple rounds of tweaking both the spec and the code later and I sent it to the dev/test environment.

I'm still not 100% on board with building the spec to perfection and only then having the model build the code. Iterating helps me really understand what the software does, especially when it includes UI elements. So why have Claude write a spec? Because tomorrow, when I resume work on this feature, Claude can simply read the spec to understand everything we've done today, which saves tens of thousands of tokens and keeps us walking down the same path. Vibe coding, by contrast, would require that Claude reverse-engineer the feature it just build in order to understand where to go next.

The dogs are both pestering me, however, so it's time to go outside.