Your team's knowledge belongs in one place an AI can read

Most organisations have the knowledge they need to run themselves well. It is just scattered across Slack threads, half-finished Notion pages, a Confluence wiki nobody updates, three generations of SharePoint, and the heads of two senior people who left last year.

This is not a tooling problem. Every company that has bought a knowledge management product has rediscovered the same thing within eighteen months: the documents go stale, the search returns nothing useful, and people give up and ask in Slack again. The wiki dies because the cost of keeping it alive is greater than the value of having it.

What changed

The reason wikis die is not that nobody can be bothered to write things down. It is that nobody can be bothered to maintain them — to update cross-references when something changes, to merge duplicate pages, to retire stale ones, to notice when the page on customer onboarding contradicts the page on customer success. That is bookkeeping work, not knowledge work, and humans are bad at it because it is boring.

LLMs do not get bored. They will happily reread a wiki, spot the contradictions, fix the cross-references and rewrite the index, every time something new is added. The bookkeeping cost has gone from "growing faster than the value" to "close to zero". For the first time in thirty years of trying, the wiki pattern is practical.

What this changes for an AI strategy

Most AI rollouts inside companies start in the wrong place. They begin with chatbots — a wrapper around the public API of a foundation model, plugged into a search index, given a brand voice and shipped to internal users. Within a quarter, two things become obvious. The bot's answers are only as good as the documents it can read. And the documents it can read are the same scattered, stale, contradictory mess they always were.

The right thing to do first is fix the substrate. Pick the small number of areas where your organisation actually depends on accumulated knowledge — the way you scope a project, the way you evaluate a vendor, the way you onboard a new client — and put each one in a single place that an LLM is responsible for keeping coherent. The rule is simple: humans add new material; the LLM keeps the structure right.

Once that exists, every downstream AI use case gets better. Search returns something useful because the underlying pages are not contradicting each other. Chatbots stop making things up because they have a clean source. New hires ramp faster because the wiki actually answers their questions. And — the part most people miss — your senior people stop being interrupted, because the things they used to be asked privately are now in a place where the LLM can answer them.

What it does not solve

This pattern does not eliminate the need for someone to decide what should be in the knowledge base in the first place. It does not turn bad source material into good source material. And it does not remove the need for human judgement on contested questions — the LLM will surface a contradiction, but it will not tell you which side is right.

What it eliminates is the part of knowledge management that no human ever wanted to do: the relentless, low-level, never-finished task of keeping a body of documentation internally consistent as it grows. That work now happens for free. Everything that used to be downstream of "but the wiki is out of date" suddenly works.

Where to start

If this lands as a useful frame, the practical first step is small. Pick one domain — not the whole company. Onboarding, or vendor management, or your sales playbook. Move the existing content into a flat directory of markdown files. Give an LLM a brief: keep these pages internally consistent, flag contradictions, maintain an index. Run it once a week.

In a month you will have a wiki that is in better shape than the one you have now. In a quarter you will know whether the pattern is worth scaling. The cost of finding out is low, and the strategic value of getting your knowledge layer right before you build serious AI on top of it is high.

That is the case for treating knowledge management as the foundation of an AI strategy, rather than an afterthought to one.