I was poking around the swiftlang macro implementation today, and I realized that I might want to experiment with some ideas to improve compiler performance. Back in the olden days, I've played around with profiling and sometimes found surprising speedups in compilers, but it's quite a time consuming affair.
Nowadays, tools like Claude code are actually good enough to at least experiment with hacky ideas to see if they're worth further pursuing, though it's very difficult to properly work in a large repo like swiftlang without very careful tuning of instructions and infra to use finite context efficiently.
When, then, will swiftlang repos adopt some minimal agentic infrastructure?
I'm guessing that relevant core contributors have either considered it already, or have very strong opinions about allowing anything of the sort to get near the repo.
I do not expect to convince the community to go AI-native, but some basics might be nice! As an open source ecosystem, given the consensus driven approaches that come there with, I suspect half the work of adoption could be done by Claude itself, if given easy access to the forums in addition to the documentation.
I'd do it myself, but I can't afford to spend the tokens on it 