miniblog.

Wonderful post from the lead LLVM maintainer reflecting on how the project is run, the design issues in LLVM today, and opportunities to make it better:

My default assumption is that external libraries are better than what I (or an LLM) would write in a v1. The extra effort to publish a project generally signifies that the author has spent a good amount of time on the problem. I end up prompting LLMs to prefer external code.

I've realised that AGENTS.md / CLAUDE.md solve two distinct problems: * It tells the LLM the conventions of the project, avoiding follow-up prompts. * It precomputes a summary of the project, saving a bunch of searches at the start of a session.