Squares and Rectangles: Why AI Agents Can’t Be Treated Like Regular Workloads

This blog post argues that AI agents should not be treated like traditional workloads and instead require a "blended identity" model that combines agent identity with the user context. It outlines shortcomings of static, pre-scoped service accounts—scope, standing access, and attribution problems—and proposes issuing runtime-scoped credentials evaluated by policy and logged for audit to reduce blast radius and support compliance without delaying deployments.