Agents as users
Delegated identity
Delegated identity gives an agent its own identity class with a delegation chain that records which agent is acting, for which person, at what scope and until when.
Control sits with the person
When to use it
Any agent acting on a person’s behalf in a system that logs who did what.
What it prevents
Attribution collapse, where two agents share credentials and accountability is lost.
Where it stands
Four architectures recur across deployments: user-delegated, autonomous with a standing identity, hybrid orchestrated, and scoped impersonation.
This entry is our own design position. It does not cite a released product or a published paper.
Related patterns: Permission parity · Agent registry
Worked through as a deck: Can an agent use your product?