What's this about?
Today's agents often need broad accessibility or disk permissions to operate applications. Two widely shared posts call for smaller capabilities, transparent activity and separate agents that audit actions and outputs.
What exactly is a security primitive?
It is a basic enforceable building block—such as “read only this folder” or “click in this app but never approve a payment”—from which safer systems can be assembled.
What's the community saying?
Krishnan drew 400 likes and 72 replies; Mollick drew 212 likes and 48 replies.
- Full-disk and accessibility grants are much broader than most agent tasks require.
- Users need a live, understandable record of what an agent is doing.
- Independent audit agents may catch mistakes, but only if they do not share the same blind spots and permissions.