The cost of tribal knowledge

Tribal knowledge is easy to underprice because it doesn't show up as a line item anywhere. Nobody budgets for it. It just quietly taxes everything a team does, in ways that are individually small enough to shrug off and collectively enormous.

The most visible cost is onboarding. A new hire doesn't actually spend their first weeks learning to code, they spend it reconstructing context that already existed somewhere, just not anywhere they could find it. The constraint behind a strange- looking function. The reason a retry sits at three attempts instead of five. That reconstruction happens one Slack message and one hallway conversation at a time, and it happens again for every new hire, because none of it gets captured the first time either.

The second cost is departure risk, and it's the more dangerous one. When a senior engineer who's been there five years leaves, the org chart doesn't show what actually walked out the door. It's every constraint they were carrying that never made it into a document, a comment, or a ticket. The next person to touch that code isn't just less experienced. They're missing information that doesn't exist anywhere else, and there's no efficient way to get it back.

The third cost is the newest one, and it's what makes the first two urgent instead of merely annoying. AI coding agents are now the primary interface a lot of engineers use to touch a codebase, and agents inherit the tribal knowledge gap completely. They read every file in seconds and still can't ask what a constraint means, so every undocumented decision becomes a landmine an agent will eventually step on, with a commit message that looks entirely reasonable right up until it isn't.