How do teams maintain understanding of large, long-lived codebases?

I’ve been working on a tool called RepoDoc, focused on tracking and explaining how real codebases evolve.

What keeps coming up is that the hardest part isn’t generating docs or summaries. It’s preserving understanding as repositories grow in size, language mix, and history.

I’m curious how teams here handle this in practice.

Do you rely on conventions, internal tooling, documentation, or mostly social knowledge? What tends to break first as a codebase gets older?