the idea is brilliant ➕

in the preview stage, kind of opening smaller group to get feedback and understanding the server cost

good question. i guess it heavily depends on the models you use. MCP basically provides context to LLM. i've seen smart model that understand my intention quite well. also seen some model doing strange stuff, try-error many times blindly

we can help ai grow too, by sharing failed attempts or precise information.
but i really hope that result model could be retained—so we don’t have to start ai from scratch like a junior dev every time.

it feels just like junior developers—we start without enough insight or wisdom, and only grow through trial and error. over time, those insights stack up and turn into experience.

we train ai for long hours hoping it can retain and reuse what it learned.
just like how we hope senior devs’ experience can be passed down.
but what kind of memory is that, really?
and how can it be preserved and reused over time?

here is the patch of react-native-ios-utilities for you gist.github.com/Kudo/1b892e3...

the code is basically referenced from github.com/expo/expo/bl...

i saw the build errors coming from react-native-ios-utilities, maybe the better way is to create a pr to add those HEADER_SEARCH_PATHS to its podspec.

though using config-plugins to update all HEADER_SEARCH_PATHS for all xcode projects also looks good to me.

it was actually a great feature request from @bacon.bsky.social for RSC streaming. glad that it helps for AI integration

i’m glad to hear that, thanks @hirbod.dev ❤️

nice 🤩