> "Either the artefact is wide open and the wrong people see it, or it is locked down so tight the right people give up and ask in Slack." ^pain A collective-track pain about the access dimension of knowledge. Two failure modes: over-sharing (artefacts reachable by readers who should not see them, including AI agents loaded into contexts they were not vetted for) and under-sharing (artefacts reachable only by the author, defeating the point of capture). A third axis emerged with AI: which agents — human or machine — may be fed which artefacts. A wiki page with no permission boundary passes through an LLM context window to anyone who can prompt the agent. ## Discovery questions - "When you grant access to a knowledge artefact, who is making that call — and based on what?" - "Has confidential information ever ended up where it shouldn't be — through a doc, through a Slack channel, through an AI tool?" - "How do you decide which artefacts an AI assistant is allowed to read?" ^discovery-questions