Knowledge management failed on labour cost, and that is the part AI absorbs
The 2000s knowledge management failures — capture overhead, staleness, disuse — were all labour problems, which is precisely the category of work AI absorbs.
Recommending knowledge management to a partner group draws a predictable response from anyone who was working in the 2000s: we tried that, it did not work. The objection is well founded. Firms built repositories, mandated contribution, appointed committees, and the results were poor often enough that the discipline acquired a reputation it has never shed.
Taking the objection seriously, rather than talking around it, is what makes the current argument credible. The question is not whether knowledge management failed. It is what it failed on.
The three failure modes
They are consistent across accounts and they are causally linked. The administrative overhead of capturing knowledge was unsustainable, so capture depended on people doing it in time they did not have. The knowledge then went stale, because keeping it fresh was more onerous than creating it. And nobody used it, because it was stale — which removed the last incentive anyone had to maintain it.
The important observation is that all three are labour problems. Not conceptual problems, not technology problems, and not failures of executive commitment. The discipline was sound and the economics of doing it by hand were not.
Why this time is different in a specific way
Labour of that shape — reading a large corpus, drafting structured material from it, noticing what has become inconsistent or out of date, retrieving the right fragment on demand — is the category of work current AI is most reliably good at. The AI can draft the capture from material the firm already generates rather than requiring people to author it separately, flag what has gone stale, and make what has been captured usable without anyone going looking for it.
The overhead that sank knowledge management the first time is now the part that can be delegated. That is a narrower claim than “AI solves knowledge management”, and the narrowness is deliberate.
What is not yet evidenced
That AI can draft capture and flag staleness is demonstrable today. That doing so is sufficient to keep a knowledge base current in live organisational use over a period of years is not something this wiki has evidence for, and the failure modes it would exhibit are not yet known. Treat the diagnosis as well supported and the prescription as promising but unproven.
One consequence follows immediately and is covered in Define a dedicated AI-facing knowledge manager role: delegating the labour does not delegate the accountability. The related failure mode of mistaking a repository for a managed knowledge base is unchanged — see A document store is not a knowledge management system.