Very rare behavior I don't understand sounds like a bummer!

and billed less potentially? When I was debugging Anthropic caching for chef.convex.dev it was enough to try to reason about our own responses, I'd have been surprised if the debugging interface I made showed something as cached and I didn't know why

I can imagine doing this for predictability. You don't want another account's behavior making your prompts cached or not, LLMs are unpredictable enough without noisy neighbors!

Very exciting!

I love the oRPC For Learning codebase!

The typed Convex function API is still missing a bunch of tricks from tRPC I'd like to add but I can tell there's even more to learn from oRPC and TanStack Start.

I feel some very cozy Escape Velocity vibes when I look at this

Exciting, looking forward to hearing what's next!

If it's something bluesky you don't have opinions about, "build me an iOS app," for me, then that's pretty fun too! But if I'm collaborating I want code I feel I could have written, which takes some work. Generally I'm going through every line, and chatting a lot about things I want changed

If you have opinions about how things ought to be done (like setting up a new project) the something existing. It is hard to convey all those preferences up front. Pretty good at copying patterns! If you have a way your crud should be written, getting it to read an existing file helps

they say CLI tool have poor discovery compared to GUIs but they don't know that we're out here sampling the input space to discover features