A bank designs a multi-agent lending assistant. Soft information collection and document checks may run autonomously, but any recommendation to approve credit above a regulatory threshold must include a human decision before systems of record update. How should the architect specify autonomy and control loops?
Select an answer to reveal the explanation.
Short Explanation
You don't give the approval agent full autonomy over credit decisions. You set autonomy by step: let low-risk collection tasks run on their own, but require human-in-the-loop approval with override paths before any high-threshold credit commitment updates systems of record.
Full Explanation
The correct approach is to set autonomy levels by step: autonomous for low-risk collection tasks, and mandatory human-in-the-loop approval with override paths before high-threshold credit commitment updates systems of record. Logical architecture skills include specifying agent autonomy levels and human-in-the-loop workflows with approval and override paths. Differentiating low-risk autonomous steps from high-stakes human-gated commitments matches regulated lending requirements. Allowing the approval agent full autonomy to write final credit decisions for all amounts is incorrect because unrestricted autonomous writes above threshold violate the stated control requirement. Removing document checks so humans see fewer intermediate artifacts is incorrect because it weakens evidence quality and does not implement the required human gate. Hiding threshold breaches from humans to reduce operational load is incorrect because concealing breaches prevents the human control the design must provide.