What is tribal knowledge?

Tribal knowledge is know-how a group shares informally but never writes down: the undocumented “how things really work here” that lives in the team rather than the docs. It travels by word of mouth — someone shows you, warns you, or corrects you, and now you know too.

The name marks the scope. Tribal knowledge is known inside one group and invisible outside it: the next team over, the new hire, and the auditor all draw a blank. The term took hold in manufacturing and quality management and spread to engineering, ops, and DevOps — the places where the documented procedure and the real one drift furthest apart.

The difference is scope: tacit knowledge is held by one person, tribal knowledge by one group, institutional knowledge by the whole organization across time. Same family of unwritten knowledge, three sizes of container.

They also fail differently. Tacit knowledge often cannot be told — the expert can do the thing and still not be able to say it. Tribal knowledge usually could be told: ask a teammate and you get a plain answer in a minute. Nothing but the writing separates it from documentation, which is exactly why it feels so cheap to leave unwritten.

Institutional knowledge (or institutional memory) is the widest lens: everything the organization has learned, including what was once written and has since been lost, buried, or fenced off in a system no one opens. The three overlap in practice — a team habit often starts as one person's tacit judgment and, if it survives long enough, hardens into institutional lore.

The deploy step no runbook mentions. The flaky test everyone reruns without comment. Knowing who to actually ask, whatever the org chart says. The real reason the Friday freeze exists. On a shop floor: which alarm means trouble and which one the night shift waves off. None of it is documented, and all of it is load-bearing.

Because it turns people into single points of failure. The measure is the bus factor: how many people can leave before the work stops. When the deploy procedure lives in two heads, your bus factor is two — one resignation away from an outage no document can end.

It also slows everyone who was not there when the knowledge formed. Onboarding stretches from weeks to months, because ramp-up is mostly absorbing the tribal layer one conversation at a time. Incidents wait on the one engineer who knows the fix. Handovers — between shifts, vendors, re-orged teams — quietly sever the transmission line. The cost stays invisible while the group is stable; it lands the day the group changes.

Because in the moment, telling is always cheaper than writing. Explaining the step to the colleague beside you costs two minutes. Documenting it properly costs an hour of your busiest expert's day — and the confusion it would prevent lands on someone else, later. Each choice is rational. The sum is a team that runs on word of mouth.

The loop feeds itself. Docs go stale, someone gets burned following one, and the lesson the team takes away is “trust the person, not the page” — which lowers the return on writing even further. Nobody decides to build a tribal layer. It accretes, one skipped write-up at a time, and the bill arrives with the first departure.

Partly. The mechanical share transfers cleanly: the missing step goes into the runbook, the real escalation path into the on-call doc. Do that — it is cheap, and it raises the bus factor immediately.

But writing changes what it writes. The caveats — when the step does not apply, which client is the exception — get trimmed for readability, and the doc starts drifting from practice the day it is published. A portion stays unwritten on purpose: the workaround that violates the official process, the frank read on a vendor — things that are safe to say and unsafe to put in writing. Documentation shrinks the tribal layer; it never empties it.

AI raises it. A general model is trained on the public written record, so it arrives holding none of your tribal knowledge. Retrieval does not close the gap: point an AI at your wiki, your drive, your tickets, and it reaches only what got written — and the tribal layer is defined by never having been written.

The failure mode is worse than a blank. The assistant confidently serves the documented procedure — the one your team knows is wrong — because the correction only ever traveled by voice. Before AI, tribal knowledge cost you onboarding time. With AI in the loop, it caps what the tools can do, and the official-but-wrong answer ships at machine speed.

Tribal knowledge is the group-held shape of dark context: what an organization knows but never wrote down — the situated context a general AI model cannot see. The term names the social scope: known inside one group, passed by word of mouth, invisible outside it.

Dark context is broader. It also covers what one person knows and cannot put into words — tacit knowledge — and what was written once but is now lost, buried, or fenced off where no one and no model can reach it. We — Ola Möller and Andriy Zhukov, at Dark Context — coined the term by analogy to dark matter: the unseen mass that holds the visible structure together. Tribal knowledge is one region of that dark mass; the map is bigger.

What is tribal knowledge?

Tribal knowledge is know-how a group shares informally but never writes down — the undocumented “how things really work here” that lives in the team rather than the docs. It passes by word of mouth and is invisible to anyone outside the group.

What is the difference between tribal knowledge and tacit knowledge?

Scope. Tacit knowledge is held by one person and often cannot be put into words — skill and judgment. Tribal knowledge is held by one group and usually could be written down; it just never was. Institutional knowledge is the widest term: what the whole organization retains over time, including records that were written and later lost.

What are examples of tribal knowledge?

The deploy step no runbook mentions. Knowing who to actually ask, whatever the org chart says. The real reason a rule exists. The flaky test everyone reruns without comment. None of it is documented, and all of it is load-bearing.

Why is tribal knowledge risky?

It makes people single points of failure. The bus factor — how many people can leave before the work stops — measures the exposure. Onboarding slows because new hires cannot search for what was never written, and incidents wait on whoever holds the fix.

How is tribal knowledge related to dark context?

Tribal knowledge is the group-held shape of dark context: what an organization knows but never wrote down — the situated context a general AI model cannot see. Dark context is broader; it also covers person-held tacit knowledge and records that were written but are now lost or fenced off. The term was coined by Ola Möller and Andriy Zhukov at Dark Context.