Every team wiki follows the same arc: enthusiastic start, slow rot, quiet death, and a channel where people ask questions the wiki was supposed to answer. The reason is structural, and the fix arrived from an unexpected direction.
A wiki decays because writing to it is a favor and reading it is a gamble. The person who knows the answer gains nothing by writing it down; the person who needs it cannot tell whether a page from last year is still true. Every stale page teaches readers to stop trusting the wiki, every distrustful reader teaches writers to stop bothering, and the spiral only turns one way.
An AI agent flips both halves of the bad bargain. Writing stops being a favor: the agent files what it learns as a side effect of doing its work, meeting notes, decisions, reconciliations, in the same folder structure people use. And reading stops being a gamble: a page carries its author and its history, so "is this current" has an answer.
A wiki agents read before answering and write after working stays current by construction, not by discipline.
The wiki must speak an open standard (MCP) so any assistant your team uses today or next year can read and write it. A wiki wired to one vendor rots the moment you switch.
Per-page access for people AND agents, or the first mistake an agent makes in a sensitive page ends the experiment.
Every edit signed, human or machine, every version kept. Trust in content is trust in its provenance.
Memava is a team wiki built to these three requirements from day one: plain files, live editing, agents as permissioned members.
Keep reading: Memava for agencies and consultancies · Memava for people building AI agents · Memava for family offices