We also need a faster language server. Inference (e.g.[1]) is one of the greatest features of TS, but we're told not to use it.

[1]: https://trpc.io/