Supplementing his article, “Redesigning Insurance Processes for Agentic AI: Governance for the Start,” Author Chris Bassett, the founder of Green Bean Consulting Group, offers this tear sheet of questions to ask during the design stage.

Controls & Recoverability

• Will it be possible to roll back an AI Agent output, along with any resultant activity?
• Will there be a tested path for a human to override a decision?
• Will each AI Agent operate within a pre-defined scope, and can those boundaries be demonstrated?
• Will the system flag concerning patterns before they become discrete errors, and pause for human evaluation when it does?

Traceability & Evidence

• Will every AI Agent action be logged in detail?
• Will each AI Agent decision be explained by the system, if requested?
• Will human decisions be documented every time, with reasoning captured in detail?

Operating Environment

• Will workflows, accountabilities and controls be fully renewed, or will they simply be repurposed?
• Will controls be built into each process, or do they sit outside the system?
• Will surrounding processes be updated, or only the target process?
• Will there be provision for governance processes to be updated based on direct system feedback?
• Will the governance model be prepared and updated so that it survives staff turnover?
• What is the cost of system monitoring and explainability, and is this proportionate in the context of anticipated benefits?