Every company starts small enough that knowledge lives in people's heads. You know who to ask. The answers are consistent. Onboarding takes an afternoon, not a quarter.
Then headcount doubles. Then doubles again. And suddenly the question "how do we do X here?" has four different answers depending on who you ask — and three of them are out of date.
This isn't a failure of effort or culture. It's a structural problem. And it has a structural fix.
The three failure modes
Watch enough teams scale and you see the same patterns.
The single point of knowledge. One person becomes the authority on a domain — a process, a system, a customer relationship. They carry the context in their head because writing it down takes time and the work is always more urgent. When they leave, or go on vacation, or just get too busy to answer, the knowledge vanishes with them.
The graveyard wiki. Teams invest real effort in documentation: Notion pages, Confluence spaces, Google Docs with hundreds of subpages. Then the product changes, the process changes, the tooling changes. Nobody updates the docs because nobody remembers they exist until they need them. When someone does find the page, it's three versions out of date and there's no way to know which parts still apply.
The search problem. Even when documentation is accurate, people often can't find it. Not because the tools are bad, but because the person searching doesn't know which words the author used. The searcher thinks "approval process," the doc says "sign-off workflow." Miss.
Why these problems compound
Each failure mode makes the others worse. Undocumented knowledge concentrates in people. People get overwhelmed answering the same questions. They write it down once, hurriedly. Then nobody maintains what was written.
And the compounding continues: as the organization grows, the ratio of explainable institutional knowledge to explained institutional knowledge keeps widening. New team members spend more time getting up to speed. Senior people spend more time explaining the same things. The tax of scale lands on everyone.
What a working solution looks like
The core insight is that documentation can't be a separate activity from work. It needs to be a byproduct of work.
Your team is already producing knowledge constantly — in Slack threads, in meeting recordings, in pull request descriptions, in customer calls. The problem is that it lives in formats that aren't queryable, aren't structured, and aren't presented to the people who need it at the moment they need it.
A system that works has three properties:
1. It sources continuously from where knowledge already lives. Not a wiki that someone has to populate. A layer that reads your existing sources — docs, chat, meetings, code — and builds structured knowledge from them automatically.
2. It stays current. Most documentation systems have no mechanism for staleness detection. A working system knows when source material changes, flags what needs review, and regenerates what it can. The goal is documentation that's wrong for days, not months.
3. It presents knowledge at the point of need. The right information, surfaced to the right person, when they're asking the question — not buried in a search result page or requiring you to already know where to look.
The right mental model
Think of institutional knowledge less like a library (static, maintained by librarians) and more like a living graph — a structure that reflects the current state of how your company actually works, that updates when the underlying work changes, and that different people can query in different ways depending on what they need to know.
The role of people in that system shifts from "write and maintain documentation" to "ensure the knowledge graph reflects reality." That's a much smaller, more tractable job.
The companies that get this right early have a compounding advantage: every new team member gets up to speed faster, every AI agent they deploy has better context to work from, and the senior people who used to be knowledge bottlenecks can focus on the work only they can do.
That's the problem Mention is built to solve. If it's one you're running into, we'd love to show you how it works.