01
Memory returns authorized context instead of starting from zero.
02
Tool permissions define what an agent may actually change.
03
Approvals and logs preserve accountability for consequential actions.
Agents need memory — and limits
Useful agents combine context, tools, permissions and human approval.