Tribal Knowledge Is the Bottleneck

Every software organization runs on knowledge that was never written down. The constraint that stops a "simple" fix from breaking production. The reason one service calls another out of the obvious order. The processor quirk that made someone move a single line of code three years ago. None of it lives in the codebase. All of it lives in someone's head.

Chapter 2 opens on a routine onboarding conversation: a new hire, a short list of components to understand, and an engineer walking through constraints that exist nowhere else: not in the wiki, not in a comment, not in a ticket. Just memory, passed down one conversation at a time, for as long as the person who remembers it stays.

That's the condition AI agents inherit when you point them at a real codebase. An agent can read every file in a repository in seconds, but it cannot sit in on that onboarding conversation. Every constraint that only exists in someone's memory is invisible to exactly the tool teams are now betting production changes on.

The chapter maps where this knowledge actually lives, in Slack threads, tribal shorthand, the specific developers who've "just always known," and makes the case that this isn't a documentation gap you can wiki your way out of. It's a structural condition with a structural cost, and naming it precisely is where the fix starts.