AI Policy Agent

Plane: Governance Plane

Purpose

Conversational policy creation. The AI Policy Agent translates user intent into formal policies through natural dialogue. It is domain-agnostic at its core — each application provides a Domain Policy Schema that gives the Agent the domain knowledge it needs.

Responsibilities

• Conduct policy creation conversations with operators

• Understand domain context via app-provided Domain Policy Schemas

• Generate formal policy definitions from natural language intent

• Validate proposed policies via the Policy Validator before activation

• Explain existing policies in natural language (audit mode)

• Simulate “what would happen if…” scenarios

Operating Modes

Mode	Purpose	Example
Creation	Help users set up new policies	“I want robots to avoid Zone B during shift changes”
Audit	Explain current policies and simulate scenarios	“What happens if a robot runs low on battery in Zone A?”

Domain Policy Schema

Applications register schemas that the Agent uses to understand domain vocabulary:

Platform provides:              App injects:
├─ Conversational engine        ├─ Domain ontology (entities, actions)
├─ Policy syntax understanding  ├─ Domain constraints
├─ Validation logic             ├─ Available strategies
└─ Common concepts              └─ Example policies / best practices
    (devices, zones, time)

Example Schema (WES domain)

{
  "domain": "warehouse",
  "entities": ["order", "pick-task", "AMR", "rack", "zone"],
  "actions": ["pick", "place", "move", "charge", "wait"],
  "strategies": ["priority-first", "proximity-first", "load-balance"],
  "constraints": ["max-carry-weight", "zone-access-level"],
  "examples": [
    "Prioritize urgent orders over standard orders",
    "Never assign more than 3 tasks to a single AMR"
  ]
}

Conversation Flow

User: "I want cleaning to stop during office hours on weekdays"
Agent: "I understand. Let me create a policy:
        - Scope: All cleaning robots
        - Condition: Monday–Friday, 9 AM – 6 PM
        - Action: Suspend cleaning operations
        - Recovery: Resume at 6 PM, prioritize uncleaned areas

        I've validated this against your existing policies — no
        conflicts detected. Should I activate it?"
User: "Yes, but only for Floor 2"
Agent: "Updated scope to Floor 2 only. Activating now."

API Surface

Operation	Description
POST /conversations	Start a new policy conversation
POST /conversations/{id}/messages	Send a message in a conversation
GET /conversations/{id}	Get conversation history
POST /conversations/{id}/activate	Activate the proposed policy
POST /audit	Start an audit query

Dependencies

Service	Relationship
Policy Service	Reads current policies, writes activated policies
Policy Validator	Validates proposed policies before activation
LLM Backend	Powers the conversational engine

Consumed By

Consumer	Usage
Applications	Embed policy creation UIs
Operators	Create and audit policies through natural conversation