miniblog.

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.

I'm still figuring out where Copilot fits in my workflow, and I find it works really well *when I know exactly what code I want*. When I have e.g. two lines in mind, I can see if it will write them (saving me the typing), and it's trivial to validate correctness.

I wish I had a good solution to link rot on my blog posts. Sometimes I think about saving all the external pages like Google's cache does. I worry about authors who don't want their content copied though.