Oct 4, 7:00 – 9:00 AM (UTC)
Salesforce Developer Group, Dubai, UAE
Should an AI agent unilaterally refund a $500 order? Renfield currently thinks yes — we disagree. This session we build zero-trust permission boundaries for autonomous agents: what they can decide alone vs. what needs a human sign-off.
Renfield is enthusiastic. Renfield is also, currently, dangerously unsupervised. Give an autonomous agent too much permission and you don't get efficiency — you get a very expensive incident report.
This session is about zero-trust design for autonomous agents: what should Renfield be allowed to decide entirely on his own (a standard exchange), versus what absolutely needs a human ("Dr. Victor") to approve (a $500 custom cape refund)? We build the permission boundary live, covering running user, Permission Sets, and approval workflows — least-privilege design applied to a real agent.
This is one of the most important (and most skipped) parts of any real Agentforce rollout. Come for FrankenForce, stay for the "wait, my org has this exact problem" realization.
Repo: github.com/FrankenOrg/frankenforce (tag: session-10-zero-trust)
Dubai Holding Assets Management
Technology Executive · AI & Salesforce Strategy | DubaiDreamin Co-Lead | CGL
CONTACT US
