AI Agents in Knowledge Management: Where It Breaks
The answer exists in writing and nobody can find it.
The loop, as it actually runs
- Knowledge captured
- structured
- indexed
- surfaced on demand
- kept current
- gaps identified
Where it breaks
Not in the steps. In the joins between them:
- Capture depends on discipline nobody has
- Search does not match how people ask
- Documents go stale with no owner
- Nobody knows what is missing until it is needed
What that actually costs
The cost of knowledge that exists but cannot be found is paid in duplicated work and inconsistent answers. Two people solve the same problem in different weeks; three people give a customer three versions of the same policy. Neither is visible as a cost because both look like ordinary work being done. The compounding version is worse: because search does not work, people stop looking, and because people stop looking, they stop writing things down, since documentation nobody reads is obviously a waste of time. The system decays for rational reasons.
What we would build first
Surfacing on demand goes first, not capture. Every knowledge management project starts with capture and most die there, because capture depends on a discipline nobody has and produces no benefit until the corpus is large. Making the existing material findable — answering questions from what is already written, in the place people ask — produces value on day one from documents you already have, and it is the thing that makes people willing to write more. Fix retrieval, and capture starts fixing itself.
It runs inside Notion, Google Drive, Slack, Confluence, Gmail — the tools already in use.
How this one goes wrong
An agent answering from company documentation will confidently cite a document that was superseded fourteen months ago, and because the answer came from a system rather than a colleague, people treat it as authoritative and stop checking. A stale answer propagates faster than a stale wiki page ever did. The guard is that every answer names its source and that source's last-reviewed date, and that documents past a staleness threshold are excluded from grounding entirely so the agent declines rather than guesses. An honest 'I do not have a current answer' is worth more than a fluent wrong one.
The individual jobs inside this loop — what starts each one and where it has to hand back to a person — are covered under admin & contracts.
Where it lands hardest: staffing & recruiting agencies, property management, insurance brokerages, managed service providers and mortgage brokerage. Each of those pages carries the sector’s own economics — the three roles that absorb the work, and the season that decides when a build can start.
In most companies this work sits with executive assistants, administrative assistants, data entry clerks, management analysts and document specialists. Each of those pages carries the wage arithmetic for that role, and says plainly where the numbers do not justify the spend.