I'm surprised that they don't do an integration test in CI where they actually start the app. (Since that's all you need to catch it)
We need to get marketshare by going fast!
I've been using AI codegen for months now, but on large projects. Turns out, the productivity multiplier that agentic AI can be scales at least partially in proportion to project size. Read that again, because I don't mean "inverse proportion".
When a codebase is small, every change touches a majority of the codebase, making parallel work difficult or impossible. Once it gets large enough to have functional areas, you can have multiple tasks running at once with little or no merge conflicts.
I was giving Cursor a shot because it's the tool that's most popular at my new company. Prior to this, I was using OpenHands. I've used Claude Code quite a bit for my personal stuff, but I wanted some hands-on experience with local tooling and Cursor was the default choice.
Now that I've got this app to the point where frontend and backend concerns are separate and the interfaces are defined I'm realizing that Cursor doesn't seem to have anything approaching Claude Code's parallel subagent support. That's... limiting.
So now I get to decide if the improvement in velocity I'll get from switching to CC will offset the time it'll take me to make the change before I have a deadline to meet.
But testing new version would have been downloading the not-yet-updated working changelog.
There are ways to deal with this of course, and I'm not defending the very vibey way that claude-code is itself developed.
There's something so unnerving about the people pushing the AI frontier being sloppy about testing. I know, it's just a CLI wrapped around the AI itself, but it suggests to me that the culture around testing there isn't as tight and thorough as I'd like it to be.