Skip to content

Design: agent role authority model (recommend vs execute, guardrails) #3

Description

@gidutz

Each agent needs a declared authority scope — what it can recommend vs. execute, and under what guardrails. Real money is moving, so this is load-bearing.

Key questions:

  • Schema for authority: per-action, per-budget-band, per-entity-type?
  • Default policies per role (media buyer, campaign-setup, AdOps)
  • How are authorities composed when one agent calls another?
  • Override / approval flow when an agent wants to act outside scope
  • How is authority versioned and audited?

Deliverable: authority schema plus a worked example for each of the three roles.

Metadata

Metadata

Assignees

No one assigned

    Labels

    pillar:rolesAgent authority scope and guardrailstype:designDesign discussion

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions